SikaRippa
04-13-2005, 01:44 PM
Geforce 6800(le) - 'Black Screen' version 2.2
-------------------------------------------------------------------------
Hopefully the final release !
Many measurement results added:
- ESR-measurements fully explained
- Oscilloscope fotos of AC-current/voltage
-------------------------------------------------------------------------
FAQ:
Q0. What is 'Black Screen'?
A0: Freezing the computer (=black screen) when under a heavy load (3d-graphics activity). This is very usual 'feature' of some type Geforce 6800le or 6800 cards. If you are not sure whether you have encountered 'black screen' please read Chapter 2.
Q1: What can I do to overcome this problem ???
A1: If ever possible RMA your card !!!! In the case this is not possible the best and easiest way is to do 'Enhanced capacitor modification'. For details take a look Chapter 3, but all information you propably need can be found Figure below:
http://img69.echo.cx/img69/4061/capmod35gz.th.jpg (http://img69.echo.cx/my.php?image=capmod35gz.jpg)
Q2: My card card cannot be modified using 'Enhanced capacitor modification', anything else I can try ?
A2: Some cases replacing the capacitors with GOOD capacitors has removed BS (see figure below, details in Ch. 3.3)
http://img529.imageshack.us/img529/2959/capmodu1b1qs.th.jpg (http://img529.imageshack.us/my.php?image=capmodu1b1qs.jpg)
Q3: Want help me to make this document better ??
A3: See most wanted list at the end of the document. Especially I request information about cards NOT mentioned (models/versions) having 'Black screen'. If you such have a 'Black screen' please post following information:
- Exact model/manufacturer/version of your card
- Estimate how long it took before the black screen occured first time
- Capacitor labels for C136 and C143 (green labels, e.g "4 8 470 6E"), see Chapter 2.2 for details
-------------------------------------------------------------------------
1. Introduction
There is a well known, unfamous, feature called 'Black Screen' related to Geforce 6800(le), GF6800XT or GF6800 graphic cards. This feature means freezing the computer (=black screen) when under a heavy load (3d-graphics activity). I wonder that first good source of information about this feature was a thread in Gainward Forum, but now *them* have removed the link (pity) [1]. The only way restoring the computer is shuting power off/on again. The purpose of this document is:
- When finished, gather/provide all relevant information about this feature (later BS)
- Help to indentify which cards/models/brands are affected
- Give solutions to reduce or remove BS e.g. a new VERY easy and electrically optimal way to make the capacitor modification is presented
- Try to give a reason why BS is occured (a new theory about reason is provided)
My motivation to write is document is only curiosity and fact that I have not found a complete document about BS. The information is scattered all over the forums and I found it very difficult to figure out the whole problem. I an only a-poor-(lonesome)-GW600LE-owner and have anything relations to Gainward (later GW). However, if some member of the GW's design team will read this document it would *very* nice to comment wheather my theory about reasons lead to BS are correct (I am sure that there ARE someone in this planet that know the real reason for BS).
I emphasize that I DO not take any responsibility of making mods suggested here or wheather the information given is fully correct. This especially case with the theory about reason of BS.
2. How to identify a BS-affected card
2.1 Software methods
BS-feature is a little bit tricky to test - sometimes it comes immediately, other day it can take minutes (hours?). Anaway it is triggered when under a high 3D-graphic load caused by SOME programs, usually games. The programs that are known to cause BS are:
- Half-life 2
- Counter-Strike Source
- Doom 3
- Far Cry
- Neverwinter Nights
- FlatOut
- WoW Beta
- Medal of Honor: Pacific Assault
- Blood Rayne II demo
- 3Dmark05
- SkiAlpin Demo 2005
- GTA San Andreas
- F.E.A.R
- Oblivion
.. to name some. A 'standard' BS-test can be performed easily with 'SkiAlpin Demo 2005', since it can be downloaded free [2]. After installing demo make following test procedure (yes, I it is German but it can be easily used even knowing a single word of German..):
1. Set 'vsync = off' at the preferences of the graphic card
2. Launch the 'SkiAlpin Demo 2005'
3. Try to have ski experience all way down - if you success without BS your card is not LIKELY affected by BS
It should be noted this does not give 100 % reliability, it has been stated that cards having BS-feature can pass this test (Finnish language ![3]). But it have been tested that most of the BS-cards fail this test, and it provides an easy way to test because BS-card fails usually very quickly. In order to make things even more complex the occurance of the BS depends many other factor i.e. operating system, power etc. (see later on this document). I has been reported that F.E.A.R (demo) is even better test program i.e. some card not giving BS with 'SkiAlpin' give BS when testing F.E.A.R
2.2 Visual inspection
There are several cards/versions/brands that are affected by BS. Some versions of the cards make BS immediately, but other can stand several months without a single BS. This makes this feature very complex and inconvenient: one can have a BS-affected card 'sleeping' or waiting the warranty time ending - and can start to make BS in future ! Therefore, I would be advantageous all users that might have potentially risky card to make a short inspection about card model/version/vendor. The studing of the card is also usefull for those who (like me..) want to find final reason why some cards are affected. Thus, I kindly ask if someone could help to get answers for questios listed at the end of this document.
Following list of the (known) affected cards is based on my investigation about the reports posted by the users.
One aspect of identifying a affected card is careful investigation of the top side capacitors ! This is explained in detailed below, but my theory about affected cards is based on differences of the capacitors. Below is a figure partly shown the top part of the GF6800 reference card. The interesting capacitors are marked with 'C141', 'C136' and 'C143'. In my theory if capacitors 'C136' and 'C143' have marking codes listed below they are potentially BS-affected ! Capacitor 'C141' has also be varied between cards, but I suppose it does not have major impact to the performance (!?).
Capacitors C136 and C143
Finally I discovered the manufacturer and type of those green labelled 'C136' and 'C143' capacitors. Following information is added to verion 1.7 of this document, so there can be several wrong or unclear details here and there in this document.
These capacitors are electrolytic capacitors with hybrid cathode electrolyte (CV-EX -series) from Sanyo. Datasheet for those capacitors can be found here [13]. The datasheet is not very informative i.e. the marking codes are not explained and therefore I made a mistake (version 1.8) when interprepting markings. I REALLY hope that this information is now correct:
- Voltage rating of 6.3 V !!
- Capacitance value 470 uF
- ESR value (max) 25 mOhm (milliohms)
- Ripple current 2090 mA
- Number 'X X' in marking code e.g. "X X 470 6E" should be lot number (usually first one is '4')
Using this information the specification for 'Ultimate capacitor modification' can be easily derivied (see chapter 3.3).
Why these capacitors would fail ? Sanyo is usually (!?) manufacturing high quality capacitors, and therefore it is quite wierd that these are not working very well. There is a comparison between different aluminium electrolytic capacitors (liquid, solid, hybrid) and their benefits and drawbacks [8]. The benefit of hybrid electrolytics is low cost at given ESR and capacitance values. Main drawback is wear out i.e. failure in long term especially at high temperatures. Good news (!) with these capacitors is that the failure mode is open i.e. even having completely 'dead' capacitor no short circuit is appeared (card can possibly be repaired by replacing capacitors). More detail about electrical characteristics are considered in other chapters.
When writing version 2.0 of this document I am fully convinced that there are versions of this card that have been altered a little bit (see Ch. 3.4). This means that these models/version are working more reliable (no BS), even when having these 'green caps'. But I am convinced that some of these capacitors ARE more poor quality, and it has been shown that the BS-problem can be removed by changing those to better capacitors (Ch 3.3). In addition to those effect, at time ofwriting version 2.1 of this document I found that a reason for 'poor quality capacitors' can be 'RoHs-compliance' (see Ch 4).
http://img106.exs.cx/img106/7294/fig12ks.th.jpg (http://img106.exs.cx/my.php?loc=img106&image=fig12ks.jpg)
Figure 1. 6800 Reference board top, important components marked
It is noteworthy to see that there are several versions for the Nvidia reference card. Below are show three different version: one having CV-EX -capacitors, second have 'unknown' capacitors and third have Sanyo SVP-capacitors.
Why they have recently (I found these images when witing version 2.0) changed the capacitor model ?
http://img355.imageshack.us/img355/3599/nv68000x19pr.th.jpg (http://img355.imageshack.us/my.php?image=nv68000x19pr.jpg)
http://img206.imageshack.us/img206/7045/nv68001x12zc.th.jpg (http://img206.imageshack.us/my.php?image=nv68001x12zc.jpg)
http://img206.imageshack.us/img206/6826/nv68002x15ae.th.jpg (http://img206.imageshack.us/my.php?image=nv68002x15ae.jpg)
Figure 2. Three different versions of the Nvidia reference cards (GF6800)
The cards that have not been affected have different markings (see figure below) like "4 2 470 6E" or even different capacitors (BGF 6800).
http://img75.exs.cx/img75/2095/fig28dw.th.jpg (http://img75.exs.cx/my.php?loc=img75&image=fig28dw.jpg)
Figure 3. Examples of 6800-cards, that not affected by 'Black Screen' (BFG6800, GW6800 and MSI GF6800)
Capacitor cross-reference
Below are listed the markings found in capacitors 'C136' and 'C143' within different 6800/6800nu/6800le cards. There is also a short explanation about what kind of BS-performance has been reported when having these capacitors.
"4 2 470 6E" - USUALLY no BS (ONE BS-case with PNY 6800 not le !!) (MSI, GW, 6800-cards),
"4 3 470 6E" - usually BS even when new (MSI, GW)
"4 4 470 6E" - no BS when new, BS after some months (GW6800le-card from RMA,XFX)
"4 6 470 6E" - can give BS when new (at least in one case) (MSI)
"4 7 470 6E" - no BS when new, BS after some months (at least some cases) (GW, Leadtek)
"4 8 470 6E" - usually no BS when new, BS after some months (MSI, GW)
"4 9 470 6E" - BS after some months ? (MSI)
Gainward
Most (?) of affected cards are manufactured by Gainward (GW). It is 99.9 % sure that there are several versions of the GW's 6800le-card and at least two that are BS-affected. Below can be found figures of the these cards that I have labeled 'version 1' and 'version 2' (see Figs. 3 and 4)
These cards look like quite same - only some components seems to be changed (see below for detailed analyze). Both cards have code of 'P64210AIP0B4", barcode of "471846200-6220", but versions 1 and 2 have manufacturing labels of 'Made in Taiwan' and 'Made in China', respectively. The main difference between these cards (in respect of BS) is that 'version 1' usually makes (made) BS right away, but 'version 2' can stand some months before the first BS. There are different capacitor markings "4 3 470 6E" for those cards "4 8 470 6E" (see figure below), which indicates that capacitors are coming from different manufactiring lot (see end of this document).
I wonder that there should GW 6800le-card that is different that explained above (even many different versions?)). I have seen some figures about GW 6800le-card having different cooler etc. , but I have not found any user report about BS-performance of this card (please post if you have any information !)
http://img106.exs.cx/img106/2677/fig36qf.th.jpg (http://img106.exs.cx/my.php?loc=img106&image=fig36qf.jpg)
Figure 4. Gainward 6800le version 1
http://img87.exs.cx/img87/770/fig44mb.th.jpg (http://img87.exs.cx/my.php?loc=img87&image=fig44mb.jpg)
Figure 5. Gainward 6800le version 2
After writing the first version I have obtained information about cards got from 'RMA'. Two cards have been reported to have capcitor markings of "4 4 470 6E" and "4 7 470 6E". See below for details differences between this model and the two 'old' versions. I also got opportunity to make electrical measurements with "4 7 470 6E" (!!!) see 'Measurement results", where can be seen that this IS really a different version of 6800le-card (some different components compared to BS-card, see Ch. 5.6). Photographs for this models are shown below.
http://img250.echo.cx/img250/4567/gw6800v1b4ni.th.jpg (http://img250.echo.cx/my.php?image=gw6800v1b4ni.jpg)
Figure 6. Gainward 6800le version 1b
Cross reference for GW-cards:
| Labels of C136 & C143 | Made in xxx | OK-Sticker | "Code 1" | "Code 3" | Version (1) | BS/Non-BS
---------------------------------------------------------------------------------------------------------------------------
| "4 3 470 6E" | Taiwan | QC-2 OK | 'missing' | G04xxxxxxx | 1 | BS ASAP
| "4 4 470 6E" | Taiwan | QC-2 OK | 'missing' | G04xxxxxxx | 1/1b (?) | BS ASAP or Non-BS ~ half year (?)
| "4 5 470 6E" | Taiwan | OK-29 | 'missing ? | G04xxxxxxx ? | 1b | BS ASAP/after few months
| "4 7 470 6E" | China | OK-188/OK-34 (3) |GW584xxxxxxx | B04xxxxxxx | 1b | Non-BS so far ??? (2)(4)
| "4 8 470 6E" | China | OK-162/OK-66 (3) |GW684xxxxxxx | B04xxxxxxx | 2 | BS after few months
| "4 9 470 6E" | China | OK-64 |CW7xxxxxxxxx | B04xxxxxxx | 1b | Non-BS so far ??? (2)
(1) Version numbering does not indicate any codes of the cards, but given by me (!)
(2) No BS-feature using standard clock frequencies at least during some months
(3) Codes varies (please post different numbers and 'features' with those models/versions)
(4) A single case of BS reported. Please post if you have "4 7 470 6E" card that is BS-affected !!
A figure for helping to identify GW6800le-cards is shown below. The most easy way to identify card is to look label of capacitors (C136 or C143) and code-labels.
http://img234.imageshack.us/img234/5371/gw6800leindentification18ry.th.jpg (http://img234.imageshack.us/my.php?image=gw6800leindentification18ry.jpg)
Figure 7. Gainward 6800le version identification
Gainward has also a 'double-fan' version of the 6800/68000le-cards. I found nice fotos of that card (see below) and it can been easily noticed used capacitors "4 3 470 6E". Since all other cards having these capacitors are BS-affected, it is possible/likely to have BS-features also with this model. Please, post if you have any extra information about this model.
http://img355.imageshack.us/img355/7694/gw6800dfx25zl.th.jpg (http://img355.imageshack.us/my.php?image=gw6800dfx25zl.jpg)
Figure 8. Gainward 6800 'double-fan'
MSI
There are also two MSI-versions of card which are affected and these have capacitor marking for 'c136' of "4 3 470 6E" "4 8 470 6E" just like Gainward. The BS-performance of those cards has been reported to be identical to Gainward i.e. card having "4 3 470 6E" capacitors is usually BS-affected very soon. MSI have also a version having capacitor markings of "4 9 470 6E", and this is also affected (AFAIK).
It has been found that MSI-cards (at least version "4 3 470 6E" !!) are the most hardest to fix with capacitor modifications. I suppose the reason is different values in the feedback path of the swithing regulator (see Ch. 3.4). Some users have reported that even after replaced capacitors ("Ultimate capacitor modification") the memory overclocking capability remains quite low level, say 750 MHz. In order to have better (overclocking) performance the feedback components could be changed, see Ch 3.4.
http://img250.echo.cx/img250/1121/msi6800lev14qv.th.jpg (http://img250.echo.cx/my.php?image=msi6800lev14qv.jpg)
Figure 9. MSI6800le version 1, having "4 3 470 6E"-capacitors
http://img356.imageshack.us/img356/9587/msi6800lev16ip.th.jpg (http://img356.imageshack.us/my.php?image=msi6800lev16ip.jpg)
Figure 10. MSI6800le version 2, having "4 9 470 6E"-capacitors (?)
MSI has also a 'repaired' version of 6800le-card. These cards have been obtained (at least in Finland) when RMA'ing BS-affected card to resellers. A fotographs of the 'repaired' MSI-card are shown below (credits for 'TuskaMies):
http://img381.imageshack.us/img381/1971/dsc035642sd6il.th.jpg (http://img381.imageshack.us/my.php?image=dsc035642sd6il.jpg)
http://img381.imageshack.us/img381/759/dsc035705hq1yl.th.jpg (http://img381.imageshack.us/my.php?image=dsc035705hq1yl.jpg)
Figure 11. MSI6800le 'repaired afterwards by manufacturer'
The reason I think this version is 'repaired afterwards by manufacturer' is poor/different soldering of C136/C143 capacitors (can be 'clearly' seen the zoomed foto). Therefore, I suppose these cards are returned (RMA'd) to MSI and 'green' CV-EX capacitors have been replaced with 'purple' SVP-capacitors. SVP-capacitors have MUCH better electrical performance these cards should be lower propability to give 'black screen. However, It can been reported (credits for 'xan') that even this version CAN GIVE BS in some cases!! I think that is due to feedback component values (Ch. 3.4) and/or 'RoHs-issues' (Ch. 4.1)
Leadtek
A few Leadtek 6800-cards have been reported to be BS-affected:
- 6800nu using "4 9 470 6E"-capacitors
- 6800le using "4 6 470 6E" or "4 7 470 6E"-capacitors
Many Leadtek card having "4 2 470 6E"-capacitors have some different issues like freezing, but no BS (?). Leadtek cards are using the same PCB and a fotograph of the 6800-card is shown below.
http://img355.imageshack.us/img355/2499/leadtek6800x14cf.th.jpg (http://img355.imageshack.us/my.php?image=leadtek6800x14cf.jpg)
Figure 12. Leadtek 6800-card
XFX
At least one XFX 6800le-card has been reported to be BS-affected, capacitor markings are not yet known (see figure below, too low quality). Please post if any extra information about XFX6800le-cards. The reason why I am VERY interesting about XFX-cards can be seen figure below, where is shown two versions of this card. It can be easily seen that these capacitors (c136, c143) HAVE BEEN CHANGED FOR THIS SECOND VERSION !!!. The zoomed foto shows that capacitor values has been increased 470 uF -> 560 uF and a good brand has been selected [12]. These capacitors have ALL features that would be advantageous in this purpose. Any ideas WHY XFX has changed these capacitors for the second model :D ? See chapter 'Ultimate capacitor modification' for details. In the time of writing version 2.1 of this document I was told that also 'version 2' of XFX6800le-card might have been BS-affected (at least one reported case by inspire'). Since, this card was fixed by replacing those capacitors the origin of the problem was the same. I wonder the reason why the (high quality SVP-capacitors) were ruined so quickly has to be related to RoHs-issues (see Chapter 4).
It has also reported that some XFX 6800XT-cards are BS-affected. These cards (GF6800XT) are almost identical to version 6800le-cards i.e. those have Sanyo SVP-capacitors for C136 and C143.
http://img194.echo.cx/img194/9025/xfx6800lev1v26ky.th.jpg (http://img194.echo.cx/my.php?image=xfx6800lev1v26ky.jpg)
Figure 13. XFX 6800le version 1 and version 2
PNY
A PNY 6800 card having "4 2 470 6E"-capacitors. A fotograph of the PNY 6800-card is shown below.
http://img185.imageshack.us/img185/8347/pny6800newold2us.th.jpg (http://img185.imageshack.us/my.php?image=pny6800newold2us.jpg)
Figure 14. PNY 6800-card
Sparkle
Sparkle 6800le/nu cards (at least some) have also been reported to be BS-affected. Credits 'xotoy' for providing foto of 6800le-card having "4 5 470 6E"-capacitors. Please, post if/when you have additional information about Sparkle-cards !
http://img235.imageshack.us/img235/5435/sparkle6800lexotoy1ci.th.jpg (http://img235.imageshack.us/my.php?image=sparkle6800lexotoy1ci.jpg)
Figure 15. Sparkle 6800le (BS-affected)
Aopen
Aopen "aeolus" 6800le cards have also been reported to be BS-affected. Credits 'jmoron' for providing foto of 6800le-card having "4 3 470 6E"-capacitors. Please, post if/when you have additional information about Aopen-cards !
http://img234.imageshack.us/img234/3485/aopen6800jmoron4xo.th.jpg (http://img234.imageshack.us/my.php?image=aopen6800jmoron4xo.jpg)
Figure 16. Aopen 6800le (BS-affected)
However, Aopen has also model (models?) that is not BS-affected at least not very soon. Credits 'sunn' for providing foto of that model (see below).
http://img235.imageshack.us/img235/1423/aopen6800lesunn9vf.th.jpg (http://img235.imageshack.us/my.php?image=aopen6800lesunn9vf.jpg)
Figure 17. Aopen 6800le (not BS-affected at least soon)
AXLE
AXLE 6800XT AGP card with "4 0 470 6E" has reported to be BS-affected. This is quite rare model, so it is not (yet) known whether all AXLE card are BS-affected.
OEM
The very first Geforce GF6800le-cards were so called OEM cards. As far as I know several manufacturers were produced those cards and some were BS-affected. At least a card made by 'Point-of-view' has been BS-affected. These cards usually (?) look like Nvidia-reference cards (see above, the first card having 'green capacitors')
Other brands
Very little information available. I have found some people complaining 'mystery' problems, hangs etc. with various 6800le-cards. It is impossible/hard to know which are BS-features. I suppose that at least some eVGA -cards are BS affected.
Please post if you have any detailed information (model/version/figure) about BS-features within these brands.
2.3 Listening noisy audio
This method sounds quite wierd, but the more I have been playing with this problem the more sure I am that mysterius sound/audio generated by 6800le-card is related to BS-problem. In order to test whether you have encounter this issue see chapter 5.4 for details. I considered that noisy audio is generated by oscillating switching regulator, and having such case makes the card more vulnerable to rapid deterioration of filtering capacitors (=could finally lead to BS). Therefore, I suggest detecting 'noisy audio' is warning signal: the card is going to give BS some day !
3. Solutions to overcome BS-feature
First and also the most important hint: try to RMA your card. This IS manufacturing problem it SHOULD be handled by the manufacturer. If you can not for some reason RMA your card, you can try some advices or modification presented below. It should be emphasized that any hardware modifiaction of your card voids the warranty immediately !
Software methods
Some software tricks can lower the propability of the BS considerably. The most easiest and maybe most efficient thing is to set 'vsync=on' at graphic card properties. At least with me this has given very good results and the games that before were totally unplayable, can with this setting keep going hours. Another thing that I noticed is the effect of operating system: using windows98 the BS probability is much more lower than with windows2000. At least when testing 'SkiAlpin' the difference between OS was huge.
It has also demostrated that antialiasing and anisotropic filtering settings have some impact to BS-performace. When having better image quality (higher AA or AF values) the probability to get BS increases.
Bios
A new version for the GW6800le bios was released after users started to complain about problems. I have not myself never tested it because it has been removed for the www-page even before I got my card. I think that this bios did not help anything - if it would, why they removed it ?? It was mentioned that cache latency for the memories were changed for that bios-version.
Hardware methods
As stated earlier, these can/will void the warranty AND some skill about electronic works is required especially when soldering !! Also be carefully about ESD when handling the card !!!
3. 1 Capacitor modification
A extremely clever modification was proposed by zadah [1] (who in my opinion is responsible of almost everything relevant electrical information related to BS-problem I have found from www, thanks & kiitti paljon!). The idea of the capacitor modification is to add an extra capacitor parallel to the capacitors 'C136' and 'C143'. Excellent figures how to make this trick can be found make locations e.g the one made by zadah [4].
The test point hole 'TP1' should be connected to the positive terminal of the capacitor and the ground of the card to the negative one (important!). The capacitor should have value ~3300 uF - 10000 (microfarads), the higher value could cause problems when the power supply of the card is starting up. The voltage rating of the capacitor should be anything well above 3 V - good values between somewhere between 6.3 V - 16 V. It has been reported (and test myself) that capacitor does not need to be soldered, but it can be pull tightly into 'TP1'. In order to make this trick working one should be sure that capacitor can not 'drop' away when the computer is running (glue tape..). Another aspect is that using capacitor modification without soldering it can be hard to get good ground contact (long wires add inductance and resistance => modification is not working). One solution (have been successfully tested!) is to solder thick (rigid) copper wire additional molex-connector and bend this wire near 'TP1' to provide support for the capacitor.
Some people have been reported that capacitor modification does not help. If you encourt such problems, please go through this check-list:
- Capacitor polarity/rating/value correct
- Is it possible to test with better capacitor, good ESR-performance (ask 'low-ESR' when buing)
- Too long wires ?
- Soldering quality
3.2 Enhanced capacitor modification
I am proud of this.. when measuring the card I found that memory voltage can be found also in 'TP10' (see Figure above). I measured the resistance between 'TP1' and 'TP10' and get value ~ 0 ohms.. since 'TP9' is ground the enhanced capacitor modification was found !
Take a look figure below to see HOW easy it put a capacitor between 'TP10' and 'TP9', even the space between those nodes is ideal. I think that 10 years child can stick a capacitor into those nodes. After I found that out it took me 2 minutes to make this modification, and it worked without soldering !!
Getting even better.., this point 'TP10' is located EXACTLY were it should be: near the inductor and MOSFET's. This improves the effect of the capacitor modification due to lower resistance and inductance of added capacitor. One more interesting feature: one can place capacitor through the PCB, so it will be located besides the other components (nice, and VERY low inductances).
I tested this modification with two capacitors having capacitance values 1000 uF and 3300 uF - both make BS disappear even without soldering !! In order to make a robust modification, I recommended soldering the leads. As a conlusion of this modification:
- EXTEREMELY easy to install - takes less than 2 minutes (if not soldered)
- Very optimal in sense of electrical properties
=> It is likely that a card before could not be modified using old strategy, can be change to working one with this modification !!! (Now, there is a proof for this [11] !!)
http://img69.echo.cx/img69/4061/capmod35gz.th.jpg (http://img69.echo.cx/my.php?image=capmod35gz.jpg)
Figure 18. Enhanced capacitor modification
If you want to optimize to performance following criteria should be taken into account when selecting capacitor:
- SMALLER ESR spec. is BETTER e.g 0.025 ohm vs 1ohm
- SMALLER dissipation factor (tan d) spec. is BETTER
- LARGER ripple current spec. is BETTER e.g. 4 A vs 1 A
- LARGER temperature spec. is BETTER e.g 105 vs 85 C
- LARGER can size spec. is BETTER (too large can be bulky)
- LARGER voltage spec. is BETTER (too large value >25 V means large can size)
3.3 Ultimate capacitor modification
It has been reported that after making capacitor modification BS's will appear again after some months. That is most likely due to fact that the c136 and c143 capacitors getting weaker all the time and the extra capacitor can not enhance the performance enough any more. Therefore, I suggest a following modification which SHOULD remove BS forever. When writing version 2.0 this modification has been successfully tested by my some others. Please, note following things before planning to do this modification:
- This modification requires some skills of soldering and decent equipment (soldering iron)
- There is a greater risk to destroy card than with other modifications
- A VERY GOOD QUALITY CAPACITORS ARE REQUIRED e.g components should beet ALL requirements listed below
For 'fast' users I made a one page modification guide (see below), but I suggest to read all hints & warnigns before trying this modidication !!!
http://img529.imageshack.us/img529/2959/capmodu1b1qs.th.jpg (http://img529.imageshack.us/my.php?image=capmodu1b1qs.jpg)
Figure 19. Ultimate capacitor modification
---
Here is a case study of card that I modified, Credits to 'money2' prodided the card for testing:
- A MSI version 1 card, with "4 3 470 6E" capacitors
- Both earlier proposed capacitor modifications have been tested, but not worked for a long time
- Without any modification maximum usable (=no BS) clock frequency was 420 MHz !!!
- After replacing "4 3 470 6E" with Chemi-con's PXA-capacitors (820 uF/6.3V) BS was removed !!
- A memory voltage could be raised at least to 814 MHz
- Case study confirms: A CARD EVEN VERY BAD CONDITION CAN BE REPAIRED USING THIS MODIFICATION
http://img5.imageshack.us/img5/6121/msi6800lev1top3004xk.th.jpg (http://img5.imageshack.us/my.php?image=msi6800lev1top3004xk.jpg)
http://img331.imageshack.us/img331/8940/msi6800lev1ctop3003rd.th.jpg (http://img331.imageshack.us/my.php?image=msi6800lev1ctop3003rd.jpg)
Figure 20. MSI version 1 before and after 'Ultimate capacitor modification'
---
The specification for the replacement capacitors are:
- ESR value less or equal than 30 mohm (milliohms)
- Acceptable ripple current MUCH greater that 2 A , preferred value ~ 4 A
- Voltage rating greater or equal than 4 V
Capacitance value greater that 470 uF, preferred value 1200 uF
- Prefer SOLID aluminium electrolytic capacitors instead of HYBRID or liquid
I am quite sure that many other vendors have proper capacitors and here are mentioned some:
- Chemi-con "PXA", PXH", "PXE" ( "PXE" is the best !!) [12]
- Sanyo "SVP" [14]
- Panasonic "WA" (not manufactured any more)
- NIC COMPONENTS CORP "NSP" [8]
- Vishay "94SVP"
- Hitano "EVS"
These capacitors might be extremely hard to find from electronic component shops. However, some worldwide distributors like 'Farnell' and 'Digi-key' provide those also in small quantities (postage& package cost might be quite high..).
I personally have used Chemicon's PXA-capacitors, but I think that the Sanyo SPV-series (used e.g. in the XFX-cards version 2) are also as-good-as PXA-capacitors. There are several capacitors listed in the reference that can be used ("10SVP560M" [12]). E.g. models "6SVP820M" and "4SVP1200M" should be fine. It should be noted very small ESR-value or different capacitance value might have negative impact to the stability of voltage regulator. I will later make electrical simulation to find out safe ranges for the values. This aspect is briefly considered below where the operation of the switching regulator is explained.
Hints & warnings
----------------
- Take care off ESD-protection i.e. use an antistatic wrap
- Do not use too much heat when desoldering and soldering. Excess heat may be harmfull for some components. Also try to work quite fast i.e. it is better to solder few times at short periods that heating the components for one long period.
- Avoid mechanical stress. When removing old components use solder sucking tool or the solder wick so you do not need to bend components
- Try to keep capacitors clean i.e. touching capacitors with oily finger tips (use rubber gloves if possible)
- DO NOT TOUCH capacitors cases or other components with a hot soldering iron
Modification
------------
1. Measure resistance between TP10-TP9 (optional but recommended, just for you can check any shorts (= much lower resistance) after modification)
2. Desolder capacitors c136 and 143 (green ones, having labels e.g "4 3 470 6E") using some solder sucking tool. I you find it hard to remove capacitors, You can GENTLY (!!) lift old capacitor with a little screwdriver etc. in the same time as heating leads. Do not throw these away, but please send me after comprehensive tests.
3. Remove old solder and any dirt under capacitors
4. Solder small amount of solder to solder pads beforehand (to help solder surface mount components)
5. Carefully solder capacitors
- Be sure to have correct polarity. There is usually a clear polarity marking (=colored line/segment for minus-terminal) in alauminum electrolytic capacitors like. Sanyo CV-EX and Chemicon's PXA
- Be sure not to have so called 'cold' solders (solder not completely melted during soldering)
6. Measure resistance between TP10-TP9 again. If you get totally different value i.e. almost 0 ohms there is a problem (=usually short made by solder under capacitor) somewere. In that case DO not install card, before you have fixed the problem !! In the case of short you should remove caps again and start the work again.
Since, I have done the modification myself I do not request anymore any tasks to be done (as I asked before). But if you have performed the mod, please post how did you succeeded:
- Card model
- Situation before (how bad BS)
- What capacitor you used
- Situation after (any problems?)
3.4 Modification of Switching regulator Feedback (used to be "Gainward modification"
I changed the name of this chapter, since it was found the the feedback component values of the switching regulator vary between cards i.e. this 'modification' is most likely not invented by Gainward - just copied to make the cards working OK !
There are GF6800le-cards that differs considerably from each other is respect of 'black screen' or more precisely the so called 'BS-frequency' is varying a lot. Here, BS-frequency means the maximum memory clock frequency that can be used WITHOUT getting BS at any circumstances. A typical BS-affected card has BS-frequency of LOWER that at stock (700 MHz), which can not be tolerated. However, it was found that with some cards the BS-frequency did NOT increase very much after ANY modification while the others are capable of running MUCH higher frequency, say 850 MHz.
A detailed analysis with cards has shown that the difference of the performance can be explained with different values of components at switching regulator feedback. I have measured myself component values of two different cards MSI (BS-affected) aand GW (no-yet, BS-affected) using undirect methods (see detail at end of document). Values of the Leadtek 6800le-card were measured by 'zxcv' (he also made nice exprerients by changing values and got HUGE boost to BS-frequency, see post #382).
Below are listed measured values:
1. C1=82n, C2=15n, C3=135n (Gainward 6800le version 1b)
2. C1=40n, C2=30n, C3=120n (MSI 6800le version 1, by 'money2')
3. C1=47n, C2=33n, C3=22n (Leadtek 6800le version "4 6 470 6E, by 'zxcv')
4. C1=100n, C2=10n, C3=10n (Leadtek 6800le version "4 6 470 6E, OPTIMAL NEW VALUES !!! by 'zxcv')
The capacitor labels are related figure below
http://img148.echo.cx/img148/2742/memfb17hp.th.jpg (http://img148.echo.cx/my.php?image=memfb17hp.jpg)
Figure 21. Locations for the switching regulator feedback components (memory voltage)
The difference between values is not large, BUT effect to performance is HUGE. I made some electrical simulations to test the effect of different values. The frequency responce of the switching regulator (memory part) is shown below:
http://img259.imageshack.us/img259/1429/memfbsimu08sr.th.jpg (http://img259.imageshack.us/my.php?image=memfbsimu08sr.jpg)
Figure 22. Simulated frequency response of the switching regulator with different component values
Some electrical backround is required to see what is purpose of those curves, but 'the higher (stabile) gain is obtained in higher frequency the faster is response' -> larger BS-frequency.
It was found the Gainward has used this trick (changing values) afterwards, since I found a 'version 1b'-card:
- Some components are 'misaligned' (see figure below), like someone soldered by hand
- Some (all) components are 'older' that version 2 card (this can be verified by VERY carefully inspection of the the MOSFETs since there is a manufacturing week printed :D )
- 'OK' label has smaller value than with 'version 2'
- Capacitor 'c141' is the same (?) as in the first version
http://img205.echo.cx/img205/9743/gv6800lev1bv2botzoom32my.th.jpg (http://img205.echo.cx/my.php?image=gv6800lev1bv2botzoom32my.jpg)
Figure 23. A zoomed area of the Gainward 6800le version 1b and 2 (switching regulator part)
Exactly the same kind of cards were reported by others (e.g. 'jabjab'), so it is quite clear the card I tested was not faulty but a number of cards has been modified by GW.
This modification can be also tested by other users, who are suffering low BS-frequency. However, please read following precautions before perform this modification:
- Changing of small SMD (surface mount) capacitors is RISKY and DIFFICULT !!
- If you still want to do it, practice yourself with old/broken PCB
- Buy a set (several different values) of suitable size ceramic SMD capacitors
- Measure resistance above capacitors before starting to desolder !!! After soldering new capacitor you should have the same value (=different means short/near to short)
- Use the values above (C1=82n, C2=15n, C3=135n or C1=100n, C2=10n, C3=10n) as a starting point. Test performance and use trial-error procedure to find optimal values
If you intend to perform this mod do not hesitate to ask more instructions & hints, and please post the results !!!
3.5 Memory voltage modification
It has been reported that adjusting a memory voltage has some effect of BS [1]. There is mentioned that lowering voltage a little bit 2.72 V -> 2.68 would remove BS. And after adding a capacitor modification the memory voltage can be increased some amount without getting BS. This can be usefull for testing purposes e.g. when overclocking the memory clock frequency.
A memory voltage modification has been proposed [4].It is quite hard to perform, since there is quite high risk to destroy the card when soldering leads to surface mounted components
I found a different way to alter memory voltage either upwards or downwards as explained in figure below.
http://img216.echo.cx/img216/4646/vmemmod16ir.th.jpg (http://img216.echo.cx/my.php?image=vmemmod16ir.jpg)
Figure 24. Memory voltage modification
This method provides a very easy way to modify memory voltage and you can test it even without soldering the leads ! Please comment if this modification has been proposed earlier so I can give credits (I have not seen before).
3.6 Power supply
It has been found that power supply of the PC has (considerably effect of BS-feature. As a rule of thumb: 'the better power supply you have the smaller is propability to get BS'. Here, better power supply means stabile 5 V and 12 V output voltages i.e. how good current supplying capability there is in molex-connector for the VGA-card. In some case BS-feature has disappear (=less frequent) when having a dedicated molex-line for card i.e. not other device like hard disk connected into the same line. I demonstrated the effect of the power supply with two examples.
I had a poor quality power supply and got BS very easily e.g. with SkiAlpin. After I installed much better supply the BS-feature disappered with SkiAlpin (using Win98), but did not disappear completely.
When measuring current consumption of BS-affected MSI-card (by 'money2') I found that a VERY small series resistance in the molex-line ( 25 - 50 milliohm!!) due to measurement setup. This resistance lowers effective voltage for the card little bit and the effect was HUGE. I got BS almost immeadiate, while testing non-BS card there was not difference.
Therefore, I think that power supply effect can explain (partly) the different behaviour of cards when installed to different computers. The supplies have little bit different output voltage level and this might be enough to trigger BS-feature mouch more easily.
4. Theory of BS-feature
The theory is present here based on the investigations of Zadah, Bilabong. They have posted *a lot* of material and it is quite hard to figure out what was their final counclusion of 'what causes BS, and why?'. I wonder that the theory of 'over-voltage-protection of the Hynix-merories' is still most popular (last?) theory in the forums. Please post if I have miss this aspect !
Anaway, I think that a proper theory should not conflict following facts, starting very general ones and some related to BS-feature (I am quite sure about these, comments please!):
F1: Exactly same reference PCB is used for many 6800 and 6800LE cards (6800NU) e.g by Gainward, MSI, Leadtek ..
F2. These gards (6800 and 6800LE) are using *quite* similar components, (except GPU of course, which is modified by disabling some units for LE/NU)
F3. All IC's of the switching regulator powering the memories (and GPU) are identical ((?))
F4. Different Hynix-memories are utilized (at least 2, 2.2, 2.5, 2.8 ns), BUT there is no logic between 6800 and 6800LE memory utilization i.e. some memory model would cause BS
F5. Passive components (capacitors and inductors) vary considerably between cards
F6. ONLY some 6800(LE) models have BS-feature !!!
F7. Some cards have BS-feature immediately e.g. GW6800LE 'version 1'
F8. Some cards might work for a while, say ~ 2-3 months, before BS-feature appears e.g. GW6800LE 'version 2'. (I think that MSI6800LE belongs to this category ?)
Moreover, a real theory of BS should give decent answer to following questions:
Q1: Why some models give BS and while others are working perfectly ?
Q2: What is the difference between those cards ?
Q3: Why some models work for a while ?
Q4: Why Gainward did not get the second card working ?
Q5: Why the cards are working with the capacitor mod?
Quite tricky questions, or what you think ? Can you answer ? I have a theory that *might* not be correct one BUT it would give, I think decent, answers to those questions.
At first some talk about the operation of the used power supply for the memories, please skip if you do not like electronics..
RoHs-issues
One very important aspect, RoHs-compliance, related to capacitance deterioration was mentioned by 'Doctor6000'. The RoHs-compliance means a new standard of using non hazarous material in electronics manufacturing [18]. In practice (in the case of GF6800le) this means they have used a new solder which higher melting point that regular, say 260 C degrees. This has been 'proofed' by me and others, since the desoldering of the capacitors is more difficult that usually (=solder does not melt easily). However, the 'green' capacitors are NOT RoHs-compatible according datasheet !!! The practice would mean the capacitors might have been (partially) ruined when manufacturing of the cards. Moreover, that would give a 'clean' explanation why some SVP-capacitors (MSI and XFX-cards) have been ruined, althought having a proper ripple current value. Also that would explain easily why some manufacturers do not have problems at all: they do not use Rohs-compatible solders !!!
4.1 Switching power supply for memories and GPU.
ISL6534
The supply voltages for the memories and GPU in the 6800(le)-cards are generated with a swithing regulator ISL6534 [5]. I draw a simplified schematic for the most important parts of the swithing regulator shown in Fig. 11. Please, note that only RELEVANT components have been drawn and there CAN be errors (it is not so easy to re-engineer multilayer PCB ;) ). At least plenty of capacitor are missing: there are many ceramic (1u - 10u, I suppose) capacitors placed all over the board to keep the power supplies clean.
In short: ISL6534 generates pulses driving MOSFET's and the squarewave output is soften with the coil and capacitor(s). The pulsewidth of the pulses are adjusted with feedbacks from memory and GPU voltages. So, if a larger current is required the upper MOS is adjusted to conduct longer period (the frequency is fixed to 300 kHz). Please refer to datasheet for detailed information. Some important details of the regulator:
http://img121.exs.cx/img121/8189/fig78fq.th.jpg (http://img121.exs.cx/my.php?loc=img121&image=fig78fq.jpg)
Figure 25. A simplified schematic of the switching power supply of the 6800-cards (for memories and GPU)
- Power supplies for ISL6534 VCC, VCC12 (5v, 12V) are NOT taken from the MOLEX-connector, but they are coming from AGP
- GPU-voltage is generated from 12V line from MOLEX
- Memory-voltage is generated from 5V line from MOLEX (!!)
I think that the following detail is the most important when considering BS-feature:
!!! ISL6534 DO NOT HAVE HAVE OVER CURRENT SENSING CIRCUIT AT ALL !!!
,and they are using diffenent methods to protect the IC from short circuit - and I think that is a key point of BS-feature. The short-circuit protection is explained as follows [5]
"There is no current sensing or rDS(ON) sensing or Under-
Voltage sensing on the ISL6534. However, if either Channel
1 or 2 output is shorted while active, there is a simple
detection on the error amp COMP output that implies either
Over-Current or Under-Voltage; the PGOOD pin goes low
immediately. If the condition persists for 1-2 internal clock
cycles (3-6µs at 300kHz), then ALL 3 Outputs are latched
off, requiring either a VCC or VCC12 POR to restart. The
protection was not designed to work for the case of powering
up an output into a short-circuit, and there are limitations on
detecting applied shorts."
In other words 'a short' is detected if the output voltage cannot be raised to specified level during 1-2 cycles. If 'short' is detected all outputs are switched OFF i.e. shutting down the GPU and memories. This will be considerd in detail in next chapter.
APL5331
DDR-memories require a 'VTT' and 'VREF' signals as reference levels for the input and output. These nodes either sink or source current and therefore a special regulator is required. ISL6534 can be connected in such manner that it generates those voltages, but these cards utilizes linear regulator APL5331 for that purpose [6]. This is somehow important aspect since this regulator increases the loading of the memory voltage.
HYNIX-memories
I think that all 6800(LE) cards are using Hynix-memories: HY5DU283222AF-xx [7]. There are some variations about speed, some cards have AF-22 and the others AF-28 etc. but basically they are quite the same chips (?). There is a single datasheet that specifies absolute maximum rating etc. so I think that they measure the speed of the chip and sell it 2.2 ns or 2.8 ns memory depending the results (such like CPU's).
I wonder that there is NOT any overprotection circuitry within those memories, in spite of the some posts. Or at least when having voltage levels << 3 V (the absolute maximum ratings for VDD is 3.6 V !!). Moreover, the termination voltage of those memories means that the signals are TERMINATED to that voltage VDD/2, so it does NOT mean 'chip will shut down when abnormal VTT voltage is detected'. Anybody expert with memories ? Please comment !!
Feedback-loop
After investigating GW version 1b -card (which had a manufacturing fault) I found how critical it is in respect of the stability of the switching regulator. In addition to ESR of the output capaciotrs (c136 and c143) at least following components have major effect to the performance of the regulator:
- All output capacitors for memory voltage and their ESR and ESL
- Inductor
- Capacitor and resistor values at the feedback loop (these were altered for GW version 1b..)
- Load resistance !! i.e. current consumption of the memories
--
Before I found the information about Sanyo CV-EX-series capacitors, I tried to make specifications on my own.
Using the equation in the datasheet of the switching regulator [5] it can be calculated that the ESR-value of the capacitor has to match to capacitance and inductor values of the output filter. This gives relationship of
ESR < SQRT(L/C) (ESR = Effective series resistance, L = inductance, C = capacitance)
Having L = 1uH ja C = 2*470 uF = 940 uF, gives ESR < 0 .033 ohm or ESR < 0.066 ohm/each capacitor (two in parallel). This value matches quite well to datasheet value.
--
The ripple current rating is a little bit tricky. The measured value for current consumption is between 0.6 - 2.64 A for 5V molex-connector which corresponds to power of 13.2 W. Power power can be much more higher due to the limitation of the multimeter current measuring badtwidth. If it is assumed that the power taken from molex-connector in this case is mainly consumed by memories and the efficiency of the switching power is 80 %, the maximum power consumed by memories is 0.8*13.2 W = 10.56 W. This corresponds current of ~3.9A @ 2.72V (memory voltage). The maximum power consumption of single Hynix-memory is 2W, so 8*2W is 16 W for all memories. Therefore the calculated value is at least in the same magnitude.
Peak current and voltage of the switching power supply can be approximated as [5]
I(ripple) = ((Vin-Vout)/(Fs*L))(Vout/Vin) (Vin = 5V, Vout = 2.72 V, L = 1uH, Fs = 300 kHz)
Vout(ripple) = I(ripple)*ESR (ESR = 0.030 ohm per each capacitor WHEN NEW ??)
This gives an approximation of I(ripple) ~ 4.2 A . There are two capacitors (C136 and C143) in parallel at output of the switching regulator, so each capacitor has to handle current of ~ 2.1 A When comparing this value to the specifications of Sanyo CV-EX capacitors (470 uF, 6.3V, I(Ripple) = 2.09 A !!!) it can be easily seen that the ripple current rating should be proper fo this purpose. But as explained above the approximation is more likely too low so the current rating is too tight is in normal condition. In addition to the ripple current the capacitors are charged/discharged heavily as explained above (consumption measured for molex-connector changes between 0.6 - 2.64 A). Moreover, the swithing regulator is oscillating (see 5.4), which causes VERY high varying current at frequency range between 100 Hz - 1000 kHz. I have not managed so far to estimate current value due to oscillation, but while having noisy audio at the CD-rom output the current level HAS to be remarkable (I will later try to measure it). Aluminium electrolytic capacitors are NOT designed for charging and discharging frequently. These aspects might be the key elements for the reason WHY capacitors fail as a new or in same case can withstand for a some time. The excess ripple current and/or high charge/disharge of aluminium electrolytic capacitors causes [9][10][15]. In addition to ripple current effects the likely utilization of 'RoHs-compatible' solder has a considerably effect of the capacitor life i.e. the capacitors will get old/ruined instanteously during manufacturing.
- Life time decrease
- Capacitance value decrease
- Increase of ESR-value
.. which all can be easily cause power supply to fail in hard load condition -> BS.
--
I have tried to simulate operation of the switching power supply. I have used a free simulator iSim can be downloaded here [17]. The schematic example for switching regulator can be also download from there after registration. I use component values obtained methods explained Ch. 5.6. It should be noted these values are not very accurate and the obtained results gives only insight about what is happening. Moreover, a number of parasitic components (resistances, inductances, capacitances) due to PCB etc. are not included to the simulation and these could have relatively large impact to overall performance.
I have been simulating AC-response and transient for the feedback loop of the memory switching regulator. I have some preliminary indicating, that the memory part of the regulator is vulnerable to unstabile operation !! I will post result later when a more accurate values for e.g. current consumption for molex-connector is available.
It is kindly asked to somebody measure feedback components (see Ch 5.5) for BS-affected card (different to MSI), so we (I) could verify weather the component values have been altered for Gainward version 1b card.
4.2 Black Screen
Here we are..what causes black screen ? My engineering guess in one word - capacitors. I think that following happens to make the BS appear. At a starting point there is a moderate load for the graphics card (this means e.g. when you launch the 'SkiAlpin' and in menu you start the game). Suddenly, the program start to write (or read) gigantic amount of data to the memory from GPU (you start skiing in 'SkiAlpin'). The memory chips sink as much current as they can for some clock cycles. This current is taken from the capacitors 'C136' and 'C143', since the switching regulator is running at frequency of 300 kHz - memories 350 MHz (700DDR). IF there are good capacitors having a low ESR (effective series resistance) and capacitance value high enough - no problem there is plenty of charge reserved and switching regulator will start pump more current within next pulse. But if there is *something bad* with those capacitor following might happen:
1. Memory voltage drops lower level
2. ISL6534 detects 'short'
3. ISL6534 tries to increase pulse width
4. If short is not disappeared within 1-3 (?) pulse all outputs are shut down !
5. Memories and GPU lose power
6. Black screen
There is an interesting comment in the datasheet of ISL6534 [5]:
"In general, the faster the rise time of the output current
during the short, the more current will be allowed on the
initial peak, and the better chance the COMP pin will have
a sharp rise as well. A low resistance short (#4) and a
higher output voltage (#5) both help. However, if the
current ramps too fast, then a false trip is also possible
(shutting down at a current level still within the expected
load range)"
.. so when having a fast current peak (=voltage level drops rapidly) is likely that power will shut off. After the BS ISL6534 is still running (see measurement results), but all outputs are 'off', which 'defends' the theory.
Another point of view: if the BS is causes by *over voltage*, how a over voltage is generated ? I think that only way to have a overvoltage is to increase the pulsewidth for the ISL6534, but what would be the mechanism to generate such situation ?
Moreover, if the memories were shut down BEFORE ISL6534, I think that the memory voltage should remain 'up' after shut down since there is not specified a lowest current that ISL6534 can drive (the minimum pulsewidth is ~0).When debugging this problem I thought that ISL6534 could be shut down due to the poor molex-lines i.e. the power supply voltage for ISL6534 would be too noisy. Therefore I measured VCC and VCC12 and found that those voltages are NOT taken from molex-connector, but AGP-bus ! Therefore noisy molex-line (due to switching power itself) cannot cause the shutdown of ISL6534.
So, I think (=I am sure) that BS is due to ISL6534 is shutting down the outputs due to faulty detected short. Yes, I know what you are going to ask: why some cards are working ? Well.. I have been investigated *a lot* the figures of the different cards and I found that the memory capacitors of these cards are the same - or...? As stated above I found that there are three numbers printed with green on the top of those capacitors like "4 3 470 6E". The middle value is capacitance in microfarads and '6E' stands for 6.3 V (I think so), but .. the first row has to means a lot number. This is not 100 % sure since I have not found the manufacturer (and datasheet) for those capacitors, but usually there is a lot number with that kind of capacitors. I found that BS-cards have certain lot numbers e.g. GW6800LE V1 "4 3". And more interesting thing two BS-cards from different manufactures have the same number like GW6800LE V2 and MSI6800LE "4 8".
Moreover, I have not seen any 6800(LE) other card (than BS-featured) with that lot number. Has there been a 'little' problem with those lots ??? I believe, yes. A lot means a large batch of components etc. manufactured in the same time, using the same materials, manufactured in the same factory etc. , so different lots can be totally different ! This application is very sensitive to following parameters:
- ESR (effective series resistance) since it determines the maximum available peak current
- Capacitance value (=how long capacitor can provide current, before voltage level drops certain amount)
- Response of the feedback loop (see Ch 3.4 above)
These kind of capacitors (aluminium electrolytic capacitor) are not so easy to manufacture and they are vulnerable to many aspect like: soldering, ripple current, temperature etc. Therefore, it is not unlikely that when buing such capacitors as-cheap-as-possible you can get very bad capacitors. I suppose that these lots "4 3" and "4 8" get somehow ruined - but it was not detected (or informed!!!). The first lot was totally bad since the cards having those capacitors made BS-immediately. Afterwards, manufactured (=Sanyo) fixed something with "4 8" and the cards might work even 3 months ! What can be wrong with those capacitors ? I think the initial capacitor values has to be correct - that would certainly be detected. Large ESR-value is harder to measure, and therefore that could be the reason the first GW-gards failed: capacitor are not capable to provide required current. The second lot "4 8" usually is working at first, so maybe they found the high ESR-value and did something;). However, the quality of the capacitors is so low that that something happens again when time goes on. A nice diagram about aluminium electrolytic capacitor failure mechanism can be found here [9]. As you can see almost anything can be happened e.g. if the ripple-current exceeds the nominal value.
A very interesting aspect about temperature effect was reported in this same thread (see posts #3, #6, #7, Thank you ZeroHero !). It was found that by cooling the ambient of the card ~ 10 degrees causes immediately BS under suitable conditions. This phenomena can be caused by many factors e.g. cooling of the switching transistors. However, the on-resistance of the transistors reduces when temperature is lower and the operation of the switching power should be better ! On the other hand the ESR-value of the aluminium electrolytic capacitors INCREASES when temperature REDUCES, which LOWERS the performance of the switching power supply (See another document about the aluminium electrolytic capacitors if you are very interested [10].). In order to make this even more complex the higher resistance (ESR) causes higher voltage drop over capacitor and in the same time the power 'burn' in the resistor increases, which increases the temperature (would be nice to make electrical model for this :D ).
This theory might be possible to proof by measuring those capacitor very accurately. This would require removing a capacitor from PCB and using a high performance inpedance meter. I am voluteer to make those measurements if someone has totally burn card and he/she is willing to provide those capacitors (either BS or non-BS cards). Please post or send a private message, so we could think how to arrange this. This would also be useful in sense of finding the final failure mechanism of those cards: will it be a short (very bad !!!) or open. In the version 2.2 this has been proofed (=measured high value of ESR), see the and of the document for measurement results.
4.3 Answers for those questions
Using proposed theory *some kind of * answers can be given:
Q1,Q2: They have poor capacitors for another manufacturing lot !
Q3,Q4: Those capacitors get *old* very quickly - some parameters (ESR or capacitance) deteriorate by temperature/ripple-current etc. But when they are new no-one cannot find anything bad !
Q5: Adding a parallel capacitor, 'overrides' the bad one - it provides enough current when high ripple current is drawn by memories
5. Measurement results
I measured several parameters from the unmodified 'GW6800le version 2'. After I got oppotunity to measure version 1B card and I performed all measurements again (!) I list below all results, some of these are not so relevant but maybe someone can use those values e.g. for debugging some other problem (?). The locations of the testpoints are shown in Fig. 13, which make it more easy to locate them when measuring a running PC. Please note the locations for TP5 and TP6 ! Those TP's were changed in the first versions of this document (1.0 - 1.6), but now tables and figure below should be ok..
http://img148.echo.cx/img148/3621/6800tps28bf.th.jpg (http://img148.echo.cx/my.php?image=6800tps28bf.jpg)
Figure 26. Test point and "PHASE"-node locations in the 6800(le)-card (bottom side of PCB).
5.1 DC-current
I measured DC-current consumption of the 5V and 12 V Molex-lines for the card. The lines provides most of the power for the card, so measuring these gives a good approximate about loading state of the card. It should be noted that these measurements are not very accurate, since the current level was changing rapidly. Therefore I provide a maximum value that I could read.
DC-current [A], vsync = off
'Version 2'
case | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
-------------------------------------------------------------------------
5V | 0.60 | 0.79 | 1.68 | 2.28 | 0.97 | 1.90 | 2.64 | 2.36 |
12V | 0.79 | 1.13 | 1.33 | 1.30 | 1.18 | 1.29 | 1.60 | 1.52 |
Please note that these values are BAD due to manufacturing bug !! (see text) 'Version 1B" (higher values due to higher voltage levels!!)
case | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
-------------------------------------------------------------------------
5V | 0.94 | 1.18 | 1.50 | 2.50 | 1.90 | 2.50 | 2.89 | 3.23 |
12V | 0.89 | 1.26 | 1.33 | 1.44 | 1.50 | 1.45 | 1.68 | 1.74 |
1. Boot
2. Idle (win98)
3. SkiAlpin -demo (main menu, win98)
4. SkiAlpin -demo (in game, win98) -> no BS
5. SkiAlpin -demo (main menu, win2000)
6. SkiAlpin -demo (in game, win2000) -> BS (for version 2)
7. 3dmark03 (Game test 1)
8. AquaMark03
It can be seen that BS occured at moderate or not maximum power level (only with version 2!), which indicates the power fails rapid increase of current not heavy but stable load.
5.2 DC-voltage
I measured Voltage from TP1 - TP14 and the 5V and 12 V Molex-lines for the card before and after BS-occured.
'Version 2'
point | TP1 | TP2 | TP3 | TP4 | TP5 | TP6 | TP7 | TP8 | TP9 | TP10 | TP11 | TP12 | TP13 | TP14 | 5V | 12V |
no-BS | 2.72 | 0.00 | 1.37 | 0.00 | 0.00 | 1.36 | 3.35 | 0.60 | 0.00 | 2.72 | 1.12 | 0.61 | 3.31 | 2.72 | 5.04 | 11.89 |
bS | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.26 | 0.05 | 0.00 | 0.00 | 0.00 | 0.00 | 3.32 | 2.72 | 5.04 | 11.89 |
Please note that these values are BAD due to manufacturing bug !! (see text) 'Version 1B'
no-BS | 3.19 | 0.00 | 1.61 | 0.00 | 0.00 | 1.60 | 3.39 | 0.61 | 0.00 | 3.18 | 1.11 | 0.60 | 3.31 | 2.72 | 5.04 | 11.93 |
TP1=TP10 (memory voltage)
TP11 (GPU voltage, small resistor 1 ohm)
TP3, TP5 (VTT/VREF voltages)
It can be clearly seen that ISL6534 operate after BS, since some reference output are working, but all switch regulator outputs have been shut down.
5.2 Resistance and capacitance
I measured the resistance and capacitance between listed nodes and GND. Please, note that capacitance values starting with ~ means uncertain value (oscillation) and the accuracy of the some capacitance values is poor (multimeter cannot measure capacitance value correctly, when there is a small parallel resistor).
'Version 2'
point | TP1 | TP2 | TP3 | TP4 | TP5 | TP6 | TP7 | TP8 | TP9 | TP10 | TP11 | TP12 | TP13 | TP14 | 5V | 12V |
resistance | 101.3 | 0 | 161.3 | 0 | 0 | 527 | 852 | 851 | 0 | 101.2 | 5.6 | 545 | 5.93k | 4.86k | 4.8k | 7.8k |
capacitance | ~5.7u | 0 | 1.46u | 0 | 0 | 125n | 20.9u | 17u | 0 | 7.9u | ~150n | 11n | 87.7n | 96.5n | 710u | 780u |
'Version 1B'
point | TP1 | TP2 | TP3 | TP4 | TP5 | TP6 | TP7 | TP8 | TP9 | TP10 | TP11 | TP12 | TP13 | TP14 | 5V | 12V |
resistance | 70.5 | 0 | 111.3 | 0 | 0 | 520 | 851 | 849 | 0 | 70.6 | 3.6 | 543 | 5.94k | 4.87k | 4.65k | 7.8k |
capacitance | ~4.7u | 0 | ~0.5u | 0 | 0 | 129n | 18.7u | 15u | 0 | ~4.7u | ~100n | 7.5n | 103nn | 111n | 702u | 772u |
5.3 AC-voltage
I measured the 'phase' node and memory voltages (see page 8 [4]) of the ISL6534 regulator circuit with HP 54602A oscilloscope. Yes, it was quite tricky operation but I succeeded ! I tried to measure if the voltage level would behave abnormally before/after BS. I found that the used oscilloscope was not able to show anything important when measuring memory voltage - the accuracy (speed) was not sufficient to detect BS. In theory that would be possible to measure, but this measurement requires a *high* cost equipment. Anyway, I managed to measure phase signal and found nice 299 kHz square wave signal having amplitude between 0 - ~5 volts. The rise and the fall times of thes signal were 13ns and 3 ns, respectively. The higher risetime is due to lower gate drive for upper MOSFET (see datasheet if you are really interested [5]). The pulse width (PW) varied rapidly when a heavy load was applied. The pulse width was 54 % in the idle state and I saw values > 60 % under heavy load. However, I did not saw pulse width higher that those values when BS occured. Again, this would indicate that BS happens at medium load by very rapid peak current that cannot ((easily)) be measured.
I also measured AC-voltage of the TP10 (Vmem) with a high performance instrumentation amplifier constructed with a PGA202. This circuit provides a high accuracy adjustable gain and rejects all common mode noise, so it is ideal to measure ripple voltages. I connected measurement probe between TP10 and TP9 as shown in foto below.
http://img405.imageshack.us/img405/2267/vmeassetup01qb.th.jpg (http://img405.imageshack.us/my.php?image=vmeassetup01qb.jpg)
Figure 27. Memory ripple voltage measurement setup
The gain was adjusted G=4, so the measured values has to be divided by four to get actual ripple voltages. Below are shown two measured cases: (a) windows in idle state (b) under heavy 3d-load. It is clearly seen that ripple voltage increases in order of four under heavy load, which indicates a high ripple current (see below).
http://img405.imageshack.us/img405/8997/vripple16mp.th.jpg (http://img405.imageshack.us/my.php?image=vripple16mp.jpg)
http://img405.imageshack.us/img405/5859/vripple20bw.th.jpg (http://img405.imageshack.us/my.php?image=vripple20bw.jpg)
Figure 28. Measurered memory ripple voltage (a) windows idle (b) under heavy 3d-load
5.4 AC-current
I measured the the AC-current consumption of the molex 5V line (which provided current for the memory regulator) as follows. I inserted a small ( 50 milliohm) in series with 5 V line of the molex-connector. Above the resistor I put a instrumentation amplifier having gain of 20 to get a reading of 1 A = 1 V in the measurement device. The foto of the setup is shown below.
http://img368.imageshack.us/img368/6728/imeassetup05am.th.jpg (http://img368.imageshack.us/my.php?image=imeassetup05am.jpg)
Figure 29. Molex 5V line ripple current measurement setup
I used an oscilloscope in different conditions to see how current consumption is varying. In the first foto below (a) is shown a windows idle state, and current spikes at low frequency is detected. This condition proofes the origin of noisy audio, since the frequency lies on audio band and traces the one that can be heard !! The rest two cases show how 'ugly' the current consumption variation can be. The current consumption varies between 1 A - 4 A which is quite near the values approximated above.
http://img455.imageshack.us/img455/2995/iripple35pn.th.jpg (http://img455.imageshack.us/my.php?image=iripple35pn.jpg)
http://img294.imageshack.us/img294/8066/iripple19ah.th.jpg (http://img294.imageshack.us/my.php?image=iripple19ah.jpg)
http://img369.imageshack.us/img369/9012/iripple24hf.th.jpg (http://img369.imageshack.us/my.php?image=iripple24hf.jpg)
Figure 30. Measurered molex 5V line ripple current (a) windows idle (b)(c) under heavy 3d-load
5.6 Noisy audio measurements
I have been solving almost whole the time when using 6800le-card quite annoying audio/noise problems. This problem has been very hard to solve, and took long time to solve it or find a sensible reason. Now I am sure that problem is caused ONLY due to GW6800le-card, and therefore I will explain the 'feature' in this document. It is likely, that the same problem disturbs also other users having a BS-card and I have demostrated excatly the same behaviour with a card never (yet) shown BS-feature.
The problem is simply a annoying sound or noise, having varing frequency and amplitude. Moreover, the frequency is 'modulated' with the current framerate (!!). Usually the noise has quite low amplitude and it can be only heard when there are no other sounds. For instance a 3dmark03-benchmark is good for testing purposes, since it does not have (mostly part) sound and the framerate is varing considerably. Following setup is required for the audio test:
- Computer having a GF6800(le)-card, that is known/assumed to be BS-affected
- CD-rom having audio cable installed to soundcard/mainbourd (important!!)
- In the 'Volume control'-panel the 'CD-player Volume' is maximized
- A suitable program is launched e.g. 3dmark03 or aquamark03 (without sound)
- Headphones preferred, since the amplitude of the noise might be low
- Just listen ..
Why I am sure it is caused by GF6800 ? I have tested a lot following different options/combinations:
- GF4200 : No noise
- GW6800le version 2 vs version 1b : No difference !!! version 1b has not made BS (yet) !!!
- Win98 vs Win2000 : No difference
- Sounblaster Audigy vs integrated : No difference
- Different CD-players : Minor difference (DVD-player makes louder noise, since it consumes more current)
- Molex removed from CD : No noise
- Added 'filter' to CD-molex : Very low noise
The filter I used is constructed with wires ripped from old PC-power and a large ferrite ring (ferrite rings can also be 'found' from old electronic equipments e.g. power supplies). The molex-wires have been cut in the middle and wounded several times around the ferrite. Finally wires are soldered together again, see figure below.
http://img245.echo.cx/img245/7923/molexferrite1oi.th.jpg (http://img245.echo.cx/my.php?image=molexferrite1oi.jpg)
Figure 31. A lowpass filter for removing power supply noise due to GF6800-card
In order to have some idea what kind of sound you would except, I recorded following audio tracks (amplified a lot): 'Windows 2000 - do nothing', ' 3dmark03 GT1', 'Aquamark03' and 'SkiAlpin'. If interested, download and enjoy :D
http://www.angelfire.com/electronic2/sikaripp/6800le/win2000_idle.mp3
http://www.angelfire.com/electronic2/sikaripp/6800le/audio_3dmark03_1.mp3
http://www.angelfire.com/electronic2/sikaripp/6800le/audio_aqua03_1.mp3
http://www.angelfire.com/electronic2/sikaripp/6800le/audio_skiapline_1.mp3
Why this is happening ??? And why it is happening also for card never made a BS-feature ? Does this noise predict BS-feature in future (if not yet affected) ?!?! I have a theory and I will explain it in more detail in the next release. In short (for this release) I think that switching regulator for memories is OSCILLATING at quite low frequency and this adds noise to the molex lines (mainly 5V).The noise is coupled via CD-player to the audio cable, due to poor power supply rejection ratio at low frequencies.
I performed more audio measurements having setup as follows. I utilized case 'windows 2000 is doing anything', to have a constant current consumption i.e. oscillation at fixed frequency. I recorded noise audio in four different cases:
- No filter at molex-connector
- 12 V molex-line filtered (12 V wire wounded around ferrite)
- 5 V molex-line filtered (5 V wire wounded around ferrite)
- Both 12 V and 5 V molex-lines filtered
Afterwards I concated files together and amplified data to make 'noise' audiable. When listening constructed file (link below), it can be easily noticed the difference due to filtering options. The filtering of 5V-molex line has much bigger effect than the 12-V line, so the main source for noise is 5V-molex line. Since, this voltage is used for memory voltage generation (and 12 V for GPU voltage), I suggest oscillation of switching regulator as a mechanism for noisy audio generation.
http://www.angelfire.com/electronic2/sikaripp/6800le/win2000_idle_all_2.mp3
5.5 Temperature measurements
I have measured temperatures of the cards: 'Gainward 6800le version 1b' and MSI 6800le (BS-affected). I used a Pt100-sensor (high accuracy platinum temperature sensor) glued at top to wooden stick (see foto below), which I gently put top of each component.
Figure 33. Temperature sensor (Pt100) attached at the tip of wooden stick
The accuracy of the sensor is high, but due to poor thermal contact the measured temperatures are too optimistic (=one or two degrees too low). Results for both card can be found below:
http://img442.imageshack.us/img442/8919/gwv1bowntemp28kt.th.jpg (http://img442.imageshack.us/my.php?image=gwv1bowntemp28kt.jpg)
http://img401.imageshack.us/img401/7384/msiv1m2temp27nx.th.jpg (http://img401.imageshack.us/my.php?image=msiv1m2temp27nx.jpg)
Figure 34. Measured temperatures of GW6800le and MSI6800le-cards
It can be seen, that the temperature of the 'green' capacitors is considerably higher in BS-affected card (MSI), which makes sense since higher ESR of capacitor means higher power consumption and higher temperature.
I measured or tested effect of very low temperature to 'Gainward 6800le version 1b' -card. It should be noted that this card has never (yet) made 'black screen', and therefore it is kindly requested wheather somebody could repeat this test with BS-affected card. Detailed instructions can be found in 'Most wanted'-list at the end of this document.
I used a freezing spray especially designed for cooling/testing of electronic equipment. If somebody will do the test it is important to use that kind of freezing method to avoid shorts etc. I launched 'the best BS-program I know' SkiAlpine2005-demo having cold spray in hand. When applying a hard 3D-load for card (=skiing down the route) I simultaneously freezed those capacitors (C136 and C143). I used cold spray A LOT to make those capacitors fully icy i.e. the temparature became well below 0 C (celcius). I did not get BS even trying this trick for little period (I wonder I am the FIRST person in this planet that have desperately tried toget BS to non-BS-card..).
Another interesting thing was that the frequency of the noisy audio did not change during freeze.
This test showed that when new (=having still good capacitors) these cards are not very sensitive to temperature. In fact in the datasheet of those capacitors is stated that they should have excellent temperature characteristics (when compared to standard aluminium electrolytic capacitors). But what kind of temperature sensitivity those capacitors have when they are almost worn out (dead) ?
I tested also the 'temperature trick' for the BS-affected MSI-card. I was running a program that did not give before BS (AquaMark03) and freezed capacitors with the spray: the result was a sudden BS !!! So, this test proofed (once again) that the problem is ESR, since ESR of these capacitors is increased when temperature goes down.
5.6 Component value determination
I have been investigated and measured the components that are part of the switching regulator. The aim of this work is to understand what the high ESR value of the capacitors causes and also find out what Gainward has been modified in the feedback path for version 1b. I think that I have now values for all relevant components and know the circuit connections at required accuracy to sketch the switching power circuitry for memory voltage generation. In this point I have now considered core-voltage generation, since I do not think it is so important for this problem.
The schematic for the switching power is quite similar to shown in the datasheet of the ISL6534 (page 17). The most infortant components in respect of the transient characteristics of the power supply are:
- Mosfets
- Inductor (output filter)
- Capacitors C136 and C143 (output filter)
- Compensation circuit
The aim of the compensation circuit is to provide proper phase margin for signal i.e. prevent oscillations etc. when heavy load transients are applied. I have been trying to analyze and make electrical simulations about this cicuit, but in order to this exact component values are required. Below I decribe in very detailed a method how I measured the component values of the feedback path. In order to find out the differences between cards (=explanations why some card are working), it is asked wheather somebody could make these measurements. These measurement requires a multimeter having capacitance and resistance measurement capability. If you will perform these DO NOT measure while power is 'ON' and be carefull not to destroy your card due to ESD (static electricity discharge)!
Following instruction are related to names shown in the Figure below.
http://img148.echo.cx/img148/2742/memfb17hp.th.jpg (http://img148.echo.cx/my.php?image=memfb17hp.jpg)
Figure 33. Locations for the switching regulator feedback components (memory voltage)
Resistors
These are quite easy to measure: no tricks are required and the polarity of the measurement does not matter (+/- -terminals of the multimeter). It is asked to measure resistance between following nodes:
R(N1-VDD_MEMORY) = 10 ohm (=R3)
R(N2-N1) = ~ 1000 ohm (=R1)
R(N2-N3) = ~ 6850 ohm (=R2)
R(TP14-GND) = ~ 4870 ohm (=R6)
R(TP14-TP13) = ~ 1000 ohm (=R5)
R(VDD_MEMORY-GND) = ~ 70 ohm
Capacitors
This is *little* bit tricky. If you are aware of basic electronics you know that in general case a resistor or capacitor connected as a part of circuit is impossible to measure. In this case those resistors can be determined by the measurement results above . The main problem with the capacitors became to fact that HAVING A RESISTOR PARALLEL TO CAPACITOR DETERIORATES RESULT. I was thing this problem *little* and I though for a long that only way to measure those values would require desoldering of components. However, I discovered two tricks and now I am quite sure that values can be determined decent accuracy with simple measurements.
Trick '1' is to short e.g with little wires some nodes to cancel some capacitors and resistors (Warning again, never do this when power is on, I HARDLY suggest to do this when card is removed from PC), . As a results only a few components are effective for measurements and values can be more easily interprepted.
Trick '2' is to afterward alter/distort/correct measured capacitance values WHEN a parallel resistance exists. I did the trick as follows. I measured capacitance and resistance values between certain nodes. Then I adjusted potentiometer (variable resistor) to measured value add soldered little arbitrary value (!) capacitors parallel to this resitor. Using trial and error I found a capacitance value which gave exactly the same capacitance reading. When removing the resistor the real capacitance value can be easily measured ! At first I tried to analytically determine the effect of the parallel resistor but turned out very nonlinear phenomena, and I decided to use trial and error method.
The capacitance values should be EXACTLY the same using this method, but I found that accuracy is lower when having low parallel resistance i.e. the obtained capacitance value is very sensitive to resistance.
I measured capacitance in following cases to determine values for relevant components (Please note that the first node was allways connected to positive terminal of multimeter!):
Easy ones:
C(TP13-GND) = 108 nF
C(TP14-GND) = 117 nF
C(N4-GND) = 155 nF
Tricky ones (need to short some points e.g. little wires NO need to solder just push gently between those nodes):
C(N3-GND = 62.6 nF (both VDD_MEMORY and N2 shorted to GND)
C(N3&N4-GND) = 1.8 nF (here N3 and N4 were shorted with measurement cable AND both VDD_MEMORY and N2 shorted to GND in the same time... yes that one is *really* tricky)
C(N2-GND) = 35 nF (here N4 AND VDD_MEMORY AND N2 were shorted to GND)
Using this data I determined capacitances (see Figure above for labels) and rounded those to standard values:
C1 = 100 nF
C2 = 10 nF
C3 = 100 nF (+/- ~50 nF)
Please note that these capacitance values might not very accurate due to reasons stated above. However, I think that these are accurate enough to provide information about feedback loop dynamics (simulations are under work..)
5.7 ESR measurements
Finally, for the version 2.2, I finised the ESR measurementS for the C136 & C143-capacitors. It was a HARD project and took ~ half a year to finish !! I would kindly thank all how send me old 'green' CV-EX capacitors, so I got enough 'material' for measurements. In the beginning I show the acquired results of the ESR-measurements. Then I briefly explain the mesurement device and try to approximate the errors due to setup.
ESR-values
I measured the ESR-values (equivalent series resistance) of several C136 & C143 capacitors ripped off the black screen affected cards. Those capacitors are CV-EX series from Sanyo and the specified ESR should be lower than 30 mohm, 470 uF 6.3 V (milliohm). The main reason I was EXTREMELY interested to acquire actual values is that a value of 2 times higher than nominal is considerd as destroyed component. So, this mesurement should confirm (and it in fact CONFIRMED) the problem is really releted to capacitors out of specifications. (On the other hand the successfully performed ultimate capacitor modification confirmed that also before, but I WANT to see on my own eyes WHAT are ESR values).
Here are the measured ESR-values for "CV-EX, 470 uF 6.3 V"-capacitors ripped off from GW and MSI 6800le-cards:
| 1 | 1 | 1 | 1 | 2 | 2 | 3 |
---------------------------------------------------------
ESR [mohm] | 207 | 194 | 199 | 211 | 192 | 170 | 227 |
Capacitor codes:
1 = "4 3 470 6E"
2 = "4 8 470 6E"
3 = "4 6 470 6E"
I am sure that everyone ever measured ESR-values within milliohm accuracy, would be interested about measurement accuracy. I considered that issue below, but to verify the proposed measurement setup I characterized also known very high quality capacitors. Below are shown measured values of brand new Chemi-Con PXA/PXH-series capacitors.
| 1 | 2 | 3 | 4 | 5 |
--------------------------------------------
ESR [mohm] | 34 | 30 | 73 | 61 | 69 |
Capacitor codes, maximum specified ESR-value in parenthesis:
1 = "PXA, 1200 uF, 4V" (10 mohm)
2 = "PXA, 820 uF, 4V" (10 mohm)
3 = "PXA, 470 uF, 6.3V" (20 mohm)
4 = "PXA, 470 uF, 4V" (20 mohm)
5 = "PXH, 680 uF, 4V" (25 mohm)
Following observations can be made using this data:
- CV-EX capacitors HAVE BEEN RUINED, since ESR-value is many times larger than specified !!
- Measurement accuracy gets worse at very low ESR-values (see below)
- Mesurement results are monotonous i.e. higher series resitance gives higher value
- Reproductability of measurements is very high i.e. the same device gives the same value when measured again
- High accurate COMPARISON between components is possible to perform
Therefore, I propose the presented mesurement results show the measured CV-EX capacitors have MUCH larger ESR than specified than brand new PXA-capacitors, having specifiactions in the same magnitude of CV-EX-capacitors. I confirmed this by connecting a very high quality capacitor in series with ~ 100 mohm resistor and got a value of ~ 100 mohm larger than without series resistance. This experiment shows that even if the meter is not operating in a linear/high accuracy operation at level of ~ 0 ohms ESR, it can operate nicely in larger values.
ESR-meter
In very beginning I considered the hardest task would be obtaining proper amount of old CV-EX capacitors. However, because of helpfull people all over the world (Poland, Germany, Finland to name some..) I got a good set of capacitors. Later I found HOW hard it is to build a ESM-meter with milliohm (or 10 milliohm) range... I build two (or in fact three) different meter before I was happy with results.
The ESR-meter I build is based on clever and simply design proposed by Ray Porter [19]. Unfortunately the link to page that explained the operation does not exist anymore, but I'll briefly explain the idea. I am not going to describe detailed the operation and how-to-build such device, because I suppose that a person how NEEDS milliohm range ERS-meter do know something about electronics.
Operation
The simplified schematic of the ESR-meter is shown below. In the original design there was a 3-pole 4-way switch selecting feedback resistors to change operation between different ranges (1, 10, 100 ohm). If you are REALLY interested about details of that please send a private message. The idea is based on a negative impedance oscillator made with an operational amplifier (X2, in Figure). When the positive and negative (resistive) feedbacks are exatly the same the pulses generated by OPAMP X1 keep X2 in a continuous oscillation. The device-under-test (=capacitor) is connected between nodes C+ and C- and the ESR-value in addition to DC-resistance of L1 and R1 sets a 'reference resistance' later RREF = ESR + R(1) + R1. So, RREF is varying between 10 - 11 ohms in the case of capacitors having maximum ESR of 1 ohm.
The operational amplifier X2 gnerates pulses, which triggers oscillator (X1). If the feedback paths are in balance (=resistor ratios R2/(RREF) (VR1+R5)/R16 are equal) that oscillation is continuous. The pulses of the oscillator are detected with X3 which drives a LED. In the other words, by adjusting a proper feedback (using VR1) the ESR of the measured capacitor is detected by inspection whether the LED is blinking. The ESR value is obtained indirect manner by observing the VR1 value in that condition. A good component for VR1 is a multiturn high precision potentiometer with a proper knob, see foto below of the ESR-meter.
http://img403.imageshack.us/img403/2593/esrmetersch11nv.th.jpg (http://img403.imageshack.us/my.php?image=esrmetersch11nv.jpg)
Figure 34. A simplified schematic diagram of the ESR-meter
Construction
The schematic of the ESR-meter includes the most of the relevant information required, but I add some comment if somebody really want to build a high performance meter. Supply voltage can be almost anything, only the maximum rating of the OPAMP's is critical. Althought a high voltage generates high pulses to the capacitor under measure, so I suggest to use a lowest voltage which is specified by OPAMP. The supply voltage could be taken directly from 9 V battery, but I found the voltage variation be too high to get milliohm accuracy. Therefore, I use a linear regulator LM317 and some good capacitors to get a clean 9 V supply from a unregulated power supply. The inductance of an inductor L1 is not too critical, but the resistance above L1+R1 (should exactly 10 ohm). In the original guide was proposed to get 14 m copper wire having diameter of 0.3 mm and coil it around screw holders of the plactic box. In the model I used (~10x10 cm) that strategy gave a square shape coil inductor with ~ 45 rounds. After I finished the coil I solder it in series with a trimmer and adjusted resistance to 10 ohm. The ABSULUTE values for VR1, R5, R16 and are not critical, but the ratios. In the original design there were used values VR1=100k, R5=1Meg, R16=11k, but I used slighly different values since I got a high precision potentiometer having value of 4.7kohm. Therefore, I suggest to first select the best potentiometer you can find having value ~ 4.7 k - 100k and scale rest values to get proper ratios: R5=10*VR1, R16=(VR1+R5)/100. In the original design there was a switch changing R5 and R16 values to get different measurement ranges. Whatever values/resistances you will use a trimmers in series will be required to get EXACTLY correct ratios. The accuracy of the feedback paths (resistor ratios) sets the accuracy of the device, so if you accept lower performance you can use standard resistors. A suggested candidate for operational amplifier is TL084/TL074: required four OPAMP's in one IC and good performance.
Calibration and Measurements
After soldering all components to PCB some kind of calibration has to be performed. In the case of meter of high (!) ESR values, say 10 ohms, it is quite easy to make cabiration. A zero ohm ESR capacitor (=in respect of the range) can be easily found, making the zero level calibration an easy task. Moreover, the range adjusting can be performed with well defined resistors e.g. by connecting 4 psc 20 ohm resistor in parallel to get a 5 ohm accurate reference. When making mesurements in milliohm range things get *a little bit* complicated:
- There is no 'short', zero ohm ESR reference, since the point is to measure the best capacitors exist !!!
- Milliohm range resistors are hard (impossible ?) to find and cannot be characterized using standard equipment
- ESR of electrolytic capacitors is highly temperature dependent i.e. touching device in the same time changes results
- Effect of the connections/wiring is dramatic: 4 cm COPPER wire diameter of 0.3 mm equals resitance of ~ 10 mohm !!!
It can be easily seen that calibration and measurements have to be performed very carefully to get reliable results. I calibrated the equipment as follows. At first I set all feedback resistors to correct values with trimmers. The LED is blinking while the feedback resistance is at correct range and therefore the value has to be acquired when LED stops blinking. The knob I used shows the position at the accuracy of 1/100 per turn, which gives in theory reading resolution of 1 mohm (10 turns). I connected all capacitors in the same position by soldering to minimize the uncertainty of lead length and contact resistance. Using a high quality capacitor (Panasonic FK, 4700 uF, 10 V, ESR(max) = 18 mohm) I set the 'zero ohm level', take a look foto below about ESR-meter with capacitor under measurements.
I found it more convenient to set zero in purpose to higher level (by trimmer settings), since the device was operation more robust/linear manner when not set set exactly to zero. Moreover, I was not sure whether I will measure later a capacitor having even lower ESR and it would not be possible to get less than zero values with the meter !! This arrangement (arbitrary zero calibration) shifts the acquired values with a fixed offset resistance Roff. I approximated the offset resistance value of 50 mohm as follows. I measured two similar Panasonic FK capacitors in parallel, which halves the ESR but keeps offset resistance unchanged and Roff can be easily calculated. I obtained a value of 50 mohm, and found the result was well reproducible between different measurement sessions. I also checked the linearity of the equipment by connecting a known resistance series with measured capacitors. I constructed some low value resistors by connecting 0.3 ohm resistors in parallel and by cutting a copper wire (diameter of 0.3 mm) to proper lengths. I found that meter was operating in a very linear manner even resistor level down to 10 milliohm, which was a 4 cm legth copper wire (!!) (see foto below for setup).
http://img133.imageshack.us/img133/5999/pict0258d8kq.th.jpg (http://img133.imageshack.us/my.php?image=pict0258d8kq.jpg)
Figure 35. The ESR-meter setup, Panasonic FK capacitor in series with 30 milliohm resistor
As a conclusion of the ESR measurements: it was not a easy task, but after using hours to examining the problem and making tests I was very happy to the results obtained. It was possible to have a proper accuracy using a home made equipment, part cost of some euros. Also these measurements finally showed that the ESR-value of the old CV-EX capacitors is far away specifications - one of the my assumptions when starting to examine BS-feature.
Afterwords:
I hope someone likes this document, at least for me solving this problem has been quite challenging and even sometimes fun ! I hope that you can help to make this document even better by posting answers releted to 'Most wanted' items.
Changelog:
1.0 Created by SikaRippa (13.4.2005)
1.1 Figure links changed (14.4.2005)
1.2 Temperature effects (ch 4.2) and anti-aliasing etc. (ch 3.1) added (19.4.2005)
1.3 Little modifications (mostly typos) (21.4.2005)
1.4 Major update (30.4.2005, "Hyvää Vappua!")
- Info about many cards added
- GW6800 version 1B (explained, measurement results)
- Ultimate modification added
1.5 Chapter 'Noisy audio measurements' added (still under construction..) (11.5.2005)
1.6 Several little modifications and 'Memory voltage modification' added (17.5.2005)
1.7 Major update (13.6.2005)
- Mystery of C136 & C143 capacitors revealed !! (Ch 2.2)
- More accurate specifications for'Ultimate capacitor modification' (Ch 3.3)
- Temperature measurements Ch 5.4
- Component value determination Ch 5.5
1.8 FAQ-list added, minor changes e.g. 'Enhanced capacitor modification' figure fixed
1.9 Some updates (24.6.2005)
- Free capacitors
- Capacitor information
- Noisy audio
- Comments about switching regulator simulation
2.0 Many updates (1.9.2005)
- Fotographs of different cards
- Capacitors lists updated
- Ultimate capacitor modification/gainward modification rewriten
2.1 Many updates (3.12.2005)
- Gainward identification more precise (figure added)
- MSI cards (foto, comments)
- Sparkle (foto, comments)
- Aopen (fotos, comments)
- "Power supply"-chapter added
- "Ultimate capacitor modification"-chapter updated (figure added)
- "Modification of Switching regulator Feedback"-chapter rewritten
- "RoHs-issue"-chapter added
- "Temperature measurements" (measurement results, comments)
2.2 Many Updates (13.4.2006, yes exactly one year....)
- New information about some models
- AC-current measurements
- AC-voltage measurements
- ESR-measurements fully explained
Super-Credits:
Zadah
Credits (sorry for those I missed, I will update list later ;) ):
Bilabong, Faults, zOU, Michael Shouler,exyxxx, MAXIMUS, Terppi, BrxA, WileNius, Maxxer, AION, maaku, JTG, Mr Q, Tiny, pkuitune, Reptile, faior, Mace, mikkomaa, ZeroHero, stoneheart,dandaman, disciple, fitnessbonk, markiemrboo, sunn, Flip, oysteini, AFA-productions (for cameras and lens ;) ), DeNs, Sidolin, julianus, money2, Godefroy, hilton, Sakapo, Tormentor, Jaxradar, Möösi, zxcv,***Deimos***, xotoy, jabjab, inspire, unrealuniverse, robb-x, jromon, soodent, piaskoon, roby58, Doctor6000, uncy_chris, Firechicken, bertos, hannibal216, OnBoard, Sh0rt, xan, Graversen, Kizmo, Augu, hilton, WarPeR, Mr.Z, anzi80, Reflex86, TuskaMies, Icharus, lettubatman, DJ22, and many others..
References:
[1] http://forum.gainward.de/viewtopic.php?t=721
[2] http://www.3dgamers.com/games/skialpin2005/downloads/
[3] http://murobbs.muropaketti.com/thread.jsp?forum=21&thread=347556&msRange=15
[4] http://www.xtremesystems.org/forums/showthread.php?t=48876
[5] http://www.intersil.com/data/fn/fn9134.pdf
[6] http://www.anpec.com.tw/products/download/APL/APL5331/APL5331.htm
[7] http://www.hynix.com/datasheet/pdf/dram/HY5DU283222AF(Rev.0.7).pdf
[8] http://www.low-esr.com/electcompare.html-ssi
[9] http://www.chemi-con.com/u7002/table2.pdf
[10] http://www.nichicon-us.com/english/lib/alminium.pdf
[11] http://forum.msi.com.tw/index.php?topic=78200.new#new
[12] http://www.semicom.co.uk/manufacturer/sanyo/sanyo_oscon/oscon_products_svp.html
[13] http://www.secc.co.jp/pdf/al_e/E15.pdf
[14] http://www.chemi-con.co.jp/Welcome_e.html
[15] http://industrial.panasonic.com/www-data/pdf/ABA0000/ABA0000TE5.pdf
[16] http://www.chemi-con.co.jp/pdf/catalog/al-e1001e/al-pxa-e-050527.pdf
[17] http://www.intersil.com/cda/deviceinfo/0%2C1477%2CISL6534%2C0.html
[18] http://www.newark.com/services/rohs/documents/RohsTechManual_v2.pdf
[19] Ray Porter, "Simple ESR Meter for Electrolytics", TELEVISION Servicing Magazine January and April 1993
------------------------------------------------------------------------------
Most Wanted list:
1. Information about BS-cards not mentioned above (brand/model/codes/etc. or foto)
2. Any comments about this document, especially about presented BS-theory.
3. Some comments from those really involved with this problem, like R&D-team of Gainward. Behave like an real engineer, and give us correct explanation about the problem (if you know !?!?)
-------------------------------------------------------------------------
Hopefully the final release !
Many measurement results added:
- ESR-measurements fully explained
- Oscilloscope fotos of AC-current/voltage
-------------------------------------------------------------------------
FAQ:
Q0. What is 'Black Screen'?
A0: Freezing the computer (=black screen) when under a heavy load (3d-graphics activity). This is very usual 'feature' of some type Geforce 6800le or 6800 cards. If you are not sure whether you have encountered 'black screen' please read Chapter 2.
Q1: What can I do to overcome this problem ???
A1: If ever possible RMA your card !!!! In the case this is not possible the best and easiest way is to do 'Enhanced capacitor modification'. For details take a look Chapter 3, but all information you propably need can be found Figure below:
http://img69.echo.cx/img69/4061/capmod35gz.th.jpg (http://img69.echo.cx/my.php?image=capmod35gz.jpg)
Q2: My card card cannot be modified using 'Enhanced capacitor modification', anything else I can try ?
A2: Some cases replacing the capacitors with GOOD capacitors has removed BS (see figure below, details in Ch. 3.3)
http://img529.imageshack.us/img529/2959/capmodu1b1qs.th.jpg (http://img529.imageshack.us/my.php?image=capmodu1b1qs.jpg)
Q3: Want help me to make this document better ??
A3: See most wanted list at the end of the document. Especially I request information about cards NOT mentioned (models/versions) having 'Black screen'. If you such have a 'Black screen' please post following information:
- Exact model/manufacturer/version of your card
- Estimate how long it took before the black screen occured first time
- Capacitor labels for C136 and C143 (green labels, e.g "4 8 470 6E"), see Chapter 2.2 for details
-------------------------------------------------------------------------
1. Introduction
There is a well known, unfamous, feature called 'Black Screen' related to Geforce 6800(le), GF6800XT or GF6800 graphic cards. This feature means freezing the computer (=black screen) when under a heavy load (3d-graphics activity). I wonder that first good source of information about this feature was a thread in Gainward Forum, but now *them* have removed the link (pity) [1]. The only way restoring the computer is shuting power off/on again. The purpose of this document is:
- When finished, gather/provide all relevant information about this feature (later BS)
- Help to indentify which cards/models/brands are affected
- Give solutions to reduce or remove BS e.g. a new VERY easy and electrically optimal way to make the capacitor modification is presented
- Try to give a reason why BS is occured (a new theory about reason is provided)
My motivation to write is document is only curiosity and fact that I have not found a complete document about BS. The information is scattered all over the forums and I found it very difficult to figure out the whole problem. I an only a-poor-(lonesome)-GW600LE-owner and have anything relations to Gainward (later GW). However, if some member of the GW's design team will read this document it would *very* nice to comment wheather my theory about reasons lead to BS are correct (I am sure that there ARE someone in this planet that know the real reason for BS).
I emphasize that I DO not take any responsibility of making mods suggested here or wheather the information given is fully correct. This especially case with the theory about reason of BS.
2. How to identify a BS-affected card
2.1 Software methods
BS-feature is a little bit tricky to test - sometimes it comes immediately, other day it can take minutes (hours?). Anaway it is triggered when under a high 3D-graphic load caused by SOME programs, usually games. The programs that are known to cause BS are:
- Half-life 2
- Counter-Strike Source
- Doom 3
- Far Cry
- Neverwinter Nights
- FlatOut
- WoW Beta
- Medal of Honor: Pacific Assault
- Blood Rayne II demo
- 3Dmark05
- SkiAlpin Demo 2005
- GTA San Andreas
- F.E.A.R
- Oblivion
.. to name some. A 'standard' BS-test can be performed easily with 'SkiAlpin Demo 2005', since it can be downloaded free [2]. After installing demo make following test procedure (yes, I it is German but it can be easily used even knowing a single word of German..):
1. Set 'vsync = off' at the preferences of the graphic card
2. Launch the 'SkiAlpin Demo 2005'
3. Try to have ski experience all way down - if you success without BS your card is not LIKELY affected by BS
It should be noted this does not give 100 % reliability, it has been stated that cards having BS-feature can pass this test (Finnish language ![3]). But it have been tested that most of the BS-cards fail this test, and it provides an easy way to test because BS-card fails usually very quickly. In order to make things even more complex the occurance of the BS depends many other factor i.e. operating system, power etc. (see later on this document). I has been reported that F.E.A.R (demo) is even better test program i.e. some card not giving BS with 'SkiAlpin' give BS when testing F.E.A.R
2.2 Visual inspection
There are several cards/versions/brands that are affected by BS. Some versions of the cards make BS immediately, but other can stand several months without a single BS. This makes this feature very complex and inconvenient: one can have a BS-affected card 'sleeping' or waiting the warranty time ending - and can start to make BS in future ! Therefore, I would be advantageous all users that might have potentially risky card to make a short inspection about card model/version/vendor. The studing of the card is also usefull for those who (like me..) want to find final reason why some cards are affected. Thus, I kindly ask if someone could help to get answers for questios listed at the end of this document.
Following list of the (known) affected cards is based on my investigation about the reports posted by the users.
One aspect of identifying a affected card is careful investigation of the top side capacitors ! This is explained in detailed below, but my theory about affected cards is based on differences of the capacitors. Below is a figure partly shown the top part of the GF6800 reference card. The interesting capacitors are marked with 'C141', 'C136' and 'C143'. In my theory if capacitors 'C136' and 'C143' have marking codes listed below they are potentially BS-affected ! Capacitor 'C141' has also be varied between cards, but I suppose it does not have major impact to the performance (!?).
Capacitors C136 and C143
Finally I discovered the manufacturer and type of those green labelled 'C136' and 'C143' capacitors. Following information is added to verion 1.7 of this document, so there can be several wrong or unclear details here and there in this document.
These capacitors are electrolytic capacitors with hybrid cathode electrolyte (CV-EX -series) from Sanyo. Datasheet for those capacitors can be found here [13]. The datasheet is not very informative i.e. the marking codes are not explained and therefore I made a mistake (version 1.8) when interprepting markings. I REALLY hope that this information is now correct:
- Voltage rating of 6.3 V !!
- Capacitance value 470 uF
- ESR value (max) 25 mOhm (milliohms)
- Ripple current 2090 mA
- Number 'X X' in marking code e.g. "X X 470 6E" should be lot number (usually first one is '4')
Using this information the specification for 'Ultimate capacitor modification' can be easily derivied (see chapter 3.3).
Why these capacitors would fail ? Sanyo is usually (!?) manufacturing high quality capacitors, and therefore it is quite wierd that these are not working very well. There is a comparison between different aluminium electrolytic capacitors (liquid, solid, hybrid) and their benefits and drawbacks [8]. The benefit of hybrid electrolytics is low cost at given ESR and capacitance values. Main drawback is wear out i.e. failure in long term especially at high temperatures. Good news (!) with these capacitors is that the failure mode is open i.e. even having completely 'dead' capacitor no short circuit is appeared (card can possibly be repaired by replacing capacitors). More detail about electrical characteristics are considered in other chapters.
When writing version 2.0 of this document I am fully convinced that there are versions of this card that have been altered a little bit (see Ch. 3.4). This means that these models/version are working more reliable (no BS), even when having these 'green caps'. But I am convinced that some of these capacitors ARE more poor quality, and it has been shown that the BS-problem can be removed by changing those to better capacitors (Ch 3.3). In addition to those effect, at time ofwriting version 2.1 of this document I found that a reason for 'poor quality capacitors' can be 'RoHs-compliance' (see Ch 4).
http://img106.exs.cx/img106/7294/fig12ks.th.jpg (http://img106.exs.cx/my.php?loc=img106&image=fig12ks.jpg)
Figure 1. 6800 Reference board top, important components marked
It is noteworthy to see that there are several versions for the Nvidia reference card. Below are show three different version: one having CV-EX -capacitors, second have 'unknown' capacitors and third have Sanyo SVP-capacitors.
Why they have recently (I found these images when witing version 2.0) changed the capacitor model ?
http://img355.imageshack.us/img355/3599/nv68000x19pr.th.jpg (http://img355.imageshack.us/my.php?image=nv68000x19pr.jpg)
http://img206.imageshack.us/img206/7045/nv68001x12zc.th.jpg (http://img206.imageshack.us/my.php?image=nv68001x12zc.jpg)
http://img206.imageshack.us/img206/6826/nv68002x15ae.th.jpg (http://img206.imageshack.us/my.php?image=nv68002x15ae.jpg)
Figure 2. Three different versions of the Nvidia reference cards (GF6800)
The cards that have not been affected have different markings (see figure below) like "4 2 470 6E" or even different capacitors (BGF 6800).
http://img75.exs.cx/img75/2095/fig28dw.th.jpg (http://img75.exs.cx/my.php?loc=img75&image=fig28dw.jpg)
Figure 3. Examples of 6800-cards, that not affected by 'Black Screen' (BFG6800, GW6800 and MSI GF6800)
Capacitor cross-reference
Below are listed the markings found in capacitors 'C136' and 'C143' within different 6800/6800nu/6800le cards. There is also a short explanation about what kind of BS-performance has been reported when having these capacitors.
"4 2 470 6E" - USUALLY no BS (ONE BS-case with PNY 6800 not le !!) (MSI, GW, 6800-cards),
"4 3 470 6E" - usually BS even when new (MSI, GW)
"4 4 470 6E" - no BS when new, BS after some months (GW6800le-card from RMA,XFX)
"4 6 470 6E" - can give BS when new (at least in one case) (MSI)
"4 7 470 6E" - no BS when new, BS after some months (at least some cases) (GW, Leadtek)
"4 8 470 6E" - usually no BS when new, BS after some months (MSI, GW)
"4 9 470 6E" - BS after some months ? (MSI)
Gainward
Most (?) of affected cards are manufactured by Gainward (GW). It is 99.9 % sure that there are several versions of the GW's 6800le-card and at least two that are BS-affected. Below can be found figures of the these cards that I have labeled 'version 1' and 'version 2' (see Figs. 3 and 4)
These cards look like quite same - only some components seems to be changed (see below for detailed analyze). Both cards have code of 'P64210AIP0B4", barcode of "471846200-6220", but versions 1 and 2 have manufacturing labels of 'Made in Taiwan' and 'Made in China', respectively. The main difference between these cards (in respect of BS) is that 'version 1' usually makes (made) BS right away, but 'version 2' can stand some months before the first BS. There are different capacitor markings "4 3 470 6E" for those cards "4 8 470 6E" (see figure below), which indicates that capacitors are coming from different manufactiring lot (see end of this document).
I wonder that there should GW 6800le-card that is different that explained above (even many different versions?)). I have seen some figures about GW 6800le-card having different cooler etc. , but I have not found any user report about BS-performance of this card (please post if you have any information !)
http://img106.exs.cx/img106/2677/fig36qf.th.jpg (http://img106.exs.cx/my.php?loc=img106&image=fig36qf.jpg)
Figure 4. Gainward 6800le version 1
http://img87.exs.cx/img87/770/fig44mb.th.jpg (http://img87.exs.cx/my.php?loc=img87&image=fig44mb.jpg)
Figure 5. Gainward 6800le version 2
After writing the first version I have obtained information about cards got from 'RMA'. Two cards have been reported to have capcitor markings of "4 4 470 6E" and "4 7 470 6E". See below for details differences between this model and the two 'old' versions. I also got opportunity to make electrical measurements with "4 7 470 6E" (!!!) see 'Measurement results", where can be seen that this IS really a different version of 6800le-card (some different components compared to BS-card, see Ch. 5.6). Photographs for this models are shown below.
http://img250.echo.cx/img250/4567/gw6800v1b4ni.th.jpg (http://img250.echo.cx/my.php?image=gw6800v1b4ni.jpg)
Figure 6. Gainward 6800le version 1b
Cross reference for GW-cards:
| Labels of C136 & C143 | Made in xxx | OK-Sticker | "Code 1" | "Code 3" | Version (1) | BS/Non-BS
---------------------------------------------------------------------------------------------------------------------------
| "4 3 470 6E" | Taiwan | QC-2 OK | 'missing' | G04xxxxxxx | 1 | BS ASAP
| "4 4 470 6E" | Taiwan | QC-2 OK | 'missing' | G04xxxxxxx | 1/1b (?) | BS ASAP or Non-BS ~ half year (?)
| "4 5 470 6E" | Taiwan | OK-29 | 'missing ? | G04xxxxxxx ? | 1b | BS ASAP/after few months
| "4 7 470 6E" | China | OK-188/OK-34 (3) |GW584xxxxxxx | B04xxxxxxx | 1b | Non-BS so far ??? (2)(4)
| "4 8 470 6E" | China | OK-162/OK-66 (3) |GW684xxxxxxx | B04xxxxxxx | 2 | BS after few months
| "4 9 470 6E" | China | OK-64 |CW7xxxxxxxxx | B04xxxxxxx | 1b | Non-BS so far ??? (2)
(1) Version numbering does not indicate any codes of the cards, but given by me (!)
(2) No BS-feature using standard clock frequencies at least during some months
(3) Codes varies (please post different numbers and 'features' with those models/versions)
(4) A single case of BS reported. Please post if you have "4 7 470 6E" card that is BS-affected !!
A figure for helping to identify GW6800le-cards is shown below. The most easy way to identify card is to look label of capacitors (C136 or C143) and code-labels.
http://img234.imageshack.us/img234/5371/gw6800leindentification18ry.th.jpg (http://img234.imageshack.us/my.php?image=gw6800leindentification18ry.jpg)
Figure 7. Gainward 6800le version identification
Gainward has also a 'double-fan' version of the 6800/68000le-cards. I found nice fotos of that card (see below) and it can been easily noticed used capacitors "4 3 470 6E". Since all other cards having these capacitors are BS-affected, it is possible/likely to have BS-features also with this model. Please, post if you have any extra information about this model.
http://img355.imageshack.us/img355/7694/gw6800dfx25zl.th.jpg (http://img355.imageshack.us/my.php?image=gw6800dfx25zl.jpg)
Figure 8. Gainward 6800 'double-fan'
MSI
There are also two MSI-versions of card which are affected and these have capacitor marking for 'c136' of "4 3 470 6E" "4 8 470 6E" just like Gainward. The BS-performance of those cards has been reported to be identical to Gainward i.e. card having "4 3 470 6E" capacitors is usually BS-affected very soon. MSI have also a version having capacitor markings of "4 9 470 6E", and this is also affected (AFAIK).
It has been found that MSI-cards (at least version "4 3 470 6E" !!) are the most hardest to fix with capacitor modifications. I suppose the reason is different values in the feedback path of the swithing regulator (see Ch. 3.4). Some users have reported that even after replaced capacitors ("Ultimate capacitor modification") the memory overclocking capability remains quite low level, say 750 MHz. In order to have better (overclocking) performance the feedback components could be changed, see Ch 3.4.
http://img250.echo.cx/img250/1121/msi6800lev14qv.th.jpg (http://img250.echo.cx/my.php?image=msi6800lev14qv.jpg)
Figure 9. MSI6800le version 1, having "4 3 470 6E"-capacitors
http://img356.imageshack.us/img356/9587/msi6800lev16ip.th.jpg (http://img356.imageshack.us/my.php?image=msi6800lev16ip.jpg)
Figure 10. MSI6800le version 2, having "4 9 470 6E"-capacitors (?)
MSI has also a 'repaired' version of 6800le-card. These cards have been obtained (at least in Finland) when RMA'ing BS-affected card to resellers. A fotographs of the 'repaired' MSI-card are shown below (credits for 'TuskaMies):
http://img381.imageshack.us/img381/1971/dsc035642sd6il.th.jpg (http://img381.imageshack.us/my.php?image=dsc035642sd6il.jpg)
http://img381.imageshack.us/img381/759/dsc035705hq1yl.th.jpg (http://img381.imageshack.us/my.php?image=dsc035705hq1yl.jpg)
Figure 11. MSI6800le 'repaired afterwards by manufacturer'
The reason I think this version is 'repaired afterwards by manufacturer' is poor/different soldering of C136/C143 capacitors (can be 'clearly' seen the zoomed foto). Therefore, I suppose these cards are returned (RMA'd) to MSI and 'green' CV-EX capacitors have been replaced with 'purple' SVP-capacitors. SVP-capacitors have MUCH better electrical performance these cards should be lower propability to give 'black screen. However, It can been reported (credits for 'xan') that even this version CAN GIVE BS in some cases!! I think that is due to feedback component values (Ch. 3.4) and/or 'RoHs-issues' (Ch. 4.1)
Leadtek
A few Leadtek 6800-cards have been reported to be BS-affected:
- 6800nu using "4 9 470 6E"-capacitors
- 6800le using "4 6 470 6E" or "4 7 470 6E"-capacitors
Many Leadtek card having "4 2 470 6E"-capacitors have some different issues like freezing, but no BS (?). Leadtek cards are using the same PCB and a fotograph of the 6800-card is shown below.
http://img355.imageshack.us/img355/2499/leadtek6800x14cf.th.jpg (http://img355.imageshack.us/my.php?image=leadtek6800x14cf.jpg)
Figure 12. Leadtek 6800-card
XFX
At least one XFX 6800le-card has been reported to be BS-affected, capacitor markings are not yet known (see figure below, too low quality). Please post if any extra information about XFX6800le-cards. The reason why I am VERY interesting about XFX-cards can be seen figure below, where is shown two versions of this card. It can be easily seen that these capacitors (c136, c143) HAVE BEEN CHANGED FOR THIS SECOND VERSION !!!. The zoomed foto shows that capacitor values has been increased 470 uF -> 560 uF and a good brand has been selected [12]. These capacitors have ALL features that would be advantageous in this purpose. Any ideas WHY XFX has changed these capacitors for the second model :D ? See chapter 'Ultimate capacitor modification' for details. In the time of writing version 2.1 of this document I was told that also 'version 2' of XFX6800le-card might have been BS-affected (at least one reported case by inspire'). Since, this card was fixed by replacing those capacitors the origin of the problem was the same. I wonder the reason why the (high quality SVP-capacitors) were ruined so quickly has to be related to RoHs-issues (see Chapter 4).
It has also reported that some XFX 6800XT-cards are BS-affected. These cards (GF6800XT) are almost identical to version 6800le-cards i.e. those have Sanyo SVP-capacitors for C136 and C143.
http://img194.echo.cx/img194/9025/xfx6800lev1v26ky.th.jpg (http://img194.echo.cx/my.php?image=xfx6800lev1v26ky.jpg)
Figure 13. XFX 6800le version 1 and version 2
PNY
A PNY 6800 card having "4 2 470 6E"-capacitors. A fotograph of the PNY 6800-card is shown below.
http://img185.imageshack.us/img185/8347/pny6800newold2us.th.jpg (http://img185.imageshack.us/my.php?image=pny6800newold2us.jpg)
Figure 14. PNY 6800-card
Sparkle
Sparkle 6800le/nu cards (at least some) have also been reported to be BS-affected. Credits 'xotoy' for providing foto of 6800le-card having "4 5 470 6E"-capacitors. Please, post if/when you have additional information about Sparkle-cards !
http://img235.imageshack.us/img235/5435/sparkle6800lexotoy1ci.th.jpg (http://img235.imageshack.us/my.php?image=sparkle6800lexotoy1ci.jpg)
Figure 15. Sparkle 6800le (BS-affected)
Aopen
Aopen "aeolus" 6800le cards have also been reported to be BS-affected. Credits 'jmoron' for providing foto of 6800le-card having "4 3 470 6E"-capacitors. Please, post if/when you have additional information about Aopen-cards !
http://img234.imageshack.us/img234/3485/aopen6800jmoron4xo.th.jpg (http://img234.imageshack.us/my.php?image=aopen6800jmoron4xo.jpg)
Figure 16. Aopen 6800le (BS-affected)
However, Aopen has also model (models?) that is not BS-affected at least not very soon. Credits 'sunn' for providing foto of that model (see below).
http://img235.imageshack.us/img235/1423/aopen6800lesunn9vf.th.jpg (http://img235.imageshack.us/my.php?image=aopen6800lesunn9vf.jpg)
Figure 17. Aopen 6800le (not BS-affected at least soon)
AXLE
AXLE 6800XT AGP card with "4 0 470 6E" has reported to be BS-affected. This is quite rare model, so it is not (yet) known whether all AXLE card are BS-affected.
OEM
The very first Geforce GF6800le-cards were so called OEM cards. As far as I know several manufacturers were produced those cards and some were BS-affected. At least a card made by 'Point-of-view' has been BS-affected. These cards usually (?) look like Nvidia-reference cards (see above, the first card having 'green capacitors')
Other brands
Very little information available. I have found some people complaining 'mystery' problems, hangs etc. with various 6800le-cards. It is impossible/hard to know which are BS-features. I suppose that at least some eVGA -cards are BS affected.
Please post if you have any detailed information (model/version/figure) about BS-features within these brands.
2.3 Listening noisy audio
This method sounds quite wierd, but the more I have been playing with this problem the more sure I am that mysterius sound/audio generated by 6800le-card is related to BS-problem. In order to test whether you have encounter this issue see chapter 5.4 for details. I considered that noisy audio is generated by oscillating switching regulator, and having such case makes the card more vulnerable to rapid deterioration of filtering capacitors (=could finally lead to BS). Therefore, I suggest detecting 'noisy audio' is warning signal: the card is going to give BS some day !
3. Solutions to overcome BS-feature
First and also the most important hint: try to RMA your card. This IS manufacturing problem it SHOULD be handled by the manufacturer. If you can not for some reason RMA your card, you can try some advices or modification presented below. It should be emphasized that any hardware modifiaction of your card voids the warranty immediately !
Software methods
Some software tricks can lower the propability of the BS considerably. The most easiest and maybe most efficient thing is to set 'vsync=on' at graphic card properties. At least with me this has given very good results and the games that before were totally unplayable, can with this setting keep going hours. Another thing that I noticed is the effect of operating system: using windows98 the BS probability is much more lower than with windows2000. At least when testing 'SkiAlpin' the difference between OS was huge.
It has also demostrated that antialiasing and anisotropic filtering settings have some impact to BS-performace. When having better image quality (higher AA or AF values) the probability to get BS increases.
Bios
A new version for the GW6800le bios was released after users started to complain about problems. I have not myself never tested it because it has been removed for the www-page even before I got my card. I think that this bios did not help anything - if it would, why they removed it ?? It was mentioned that cache latency for the memories were changed for that bios-version.
Hardware methods
As stated earlier, these can/will void the warranty AND some skill about electronic works is required especially when soldering !! Also be carefully about ESD when handling the card !!!
3. 1 Capacitor modification
A extremely clever modification was proposed by zadah [1] (who in my opinion is responsible of almost everything relevant electrical information related to BS-problem I have found from www, thanks & kiitti paljon!). The idea of the capacitor modification is to add an extra capacitor parallel to the capacitors 'C136' and 'C143'. Excellent figures how to make this trick can be found make locations e.g the one made by zadah [4].
The test point hole 'TP1' should be connected to the positive terminal of the capacitor and the ground of the card to the negative one (important!). The capacitor should have value ~3300 uF - 10000 (microfarads), the higher value could cause problems when the power supply of the card is starting up. The voltage rating of the capacitor should be anything well above 3 V - good values between somewhere between 6.3 V - 16 V. It has been reported (and test myself) that capacitor does not need to be soldered, but it can be pull tightly into 'TP1'. In order to make this trick working one should be sure that capacitor can not 'drop' away when the computer is running (glue tape..). Another aspect is that using capacitor modification without soldering it can be hard to get good ground contact (long wires add inductance and resistance => modification is not working). One solution (have been successfully tested!) is to solder thick (rigid) copper wire additional molex-connector and bend this wire near 'TP1' to provide support for the capacitor.
Some people have been reported that capacitor modification does not help. If you encourt such problems, please go through this check-list:
- Capacitor polarity/rating/value correct
- Is it possible to test with better capacitor, good ESR-performance (ask 'low-ESR' when buing)
- Too long wires ?
- Soldering quality
3.2 Enhanced capacitor modification
I am proud of this.. when measuring the card I found that memory voltage can be found also in 'TP10' (see Figure above). I measured the resistance between 'TP1' and 'TP10' and get value ~ 0 ohms.. since 'TP9' is ground the enhanced capacitor modification was found !
Take a look figure below to see HOW easy it put a capacitor between 'TP10' and 'TP9', even the space between those nodes is ideal. I think that 10 years child can stick a capacitor into those nodes. After I found that out it took me 2 minutes to make this modification, and it worked without soldering !!
Getting even better.., this point 'TP10' is located EXACTLY were it should be: near the inductor and MOSFET's. This improves the effect of the capacitor modification due to lower resistance and inductance of added capacitor. One more interesting feature: one can place capacitor through the PCB, so it will be located besides the other components (nice, and VERY low inductances).
I tested this modification with two capacitors having capacitance values 1000 uF and 3300 uF - both make BS disappear even without soldering !! In order to make a robust modification, I recommended soldering the leads. As a conlusion of this modification:
- EXTEREMELY easy to install - takes less than 2 minutes (if not soldered)
- Very optimal in sense of electrical properties
=> It is likely that a card before could not be modified using old strategy, can be change to working one with this modification !!! (Now, there is a proof for this [11] !!)
http://img69.echo.cx/img69/4061/capmod35gz.th.jpg (http://img69.echo.cx/my.php?image=capmod35gz.jpg)
Figure 18. Enhanced capacitor modification
If you want to optimize to performance following criteria should be taken into account when selecting capacitor:
- SMALLER ESR spec. is BETTER e.g 0.025 ohm vs 1ohm
- SMALLER dissipation factor (tan d) spec. is BETTER
- LARGER ripple current spec. is BETTER e.g. 4 A vs 1 A
- LARGER temperature spec. is BETTER e.g 105 vs 85 C
- LARGER can size spec. is BETTER (too large can be bulky)
- LARGER voltage spec. is BETTER (too large value >25 V means large can size)
3.3 Ultimate capacitor modification
It has been reported that after making capacitor modification BS's will appear again after some months. That is most likely due to fact that the c136 and c143 capacitors getting weaker all the time and the extra capacitor can not enhance the performance enough any more. Therefore, I suggest a following modification which SHOULD remove BS forever. When writing version 2.0 this modification has been successfully tested by my some others. Please, note following things before planning to do this modification:
- This modification requires some skills of soldering and decent equipment (soldering iron)
- There is a greater risk to destroy card than with other modifications
- A VERY GOOD QUALITY CAPACITORS ARE REQUIRED e.g components should beet ALL requirements listed below
For 'fast' users I made a one page modification guide (see below), but I suggest to read all hints & warnigns before trying this modidication !!!
http://img529.imageshack.us/img529/2959/capmodu1b1qs.th.jpg (http://img529.imageshack.us/my.php?image=capmodu1b1qs.jpg)
Figure 19. Ultimate capacitor modification
---
Here is a case study of card that I modified, Credits to 'money2' prodided the card for testing:
- A MSI version 1 card, with "4 3 470 6E" capacitors
- Both earlier proposed capacitor modifications have been tested, but not worked for a long time
- Without any modification maximum usable (=no BS) clock frequency was 420 MHz !!!
- After replacing "4 3 470 6E" with Chemi-con's PXA-capacitors (820 uF/6.3V) BS was removed !!
- A memory voltage could be raised at least to 814 MHz
- Case study confirms: A CARD EVEN VERY BAD CONDITION CAN BE REPAIRED USING THIS MODIFICATION
http://img5.imageshack.us/img5/6121/msi6800lev1top3004xk.th.jpg (http://img5.imageshack.us/my.php?image=msi6800lev1top3004xk.jpg)
http://img331.imageshack.us/img331/8940/msi6800lev1ctop3003rd.th.jpg (http://img331.imageshack.us/my.php?image=msi6800lev1ctop3003rd.jpg)
Figure 20. MSI version 1 before and after 'Ultimate capacitor modification'
---
The specification for the replacement capacitors are:
- ESR value less or equal than 30 mohm (milliohms)
- Acceptable ripple current MUCH greater that 2 A , preferred value ~ 4 A
- Voltage rating greater or equal than 4 V
Capacitance value greater that 470 uF, preferred value 1200 uF
- Prefer SOLID aluminium electrolytic capacitors instead of HYBRID or liquid
I am quite sure that many other vendors have proper capacitors and here are mentioned some:
- Chemi-con "PXA", PXH", "PXE" ( "PXE" is the best !!) [12]
- Sanyo "SVP" [14]
- Panasonic "WA" (not manufactured any more)
- NIC COMPONENTS CORP "NSP" [8]
- Vishay "94SVP"
- Hitano "EVS"
These capacitors might be extremely hard to find from electronic component shops. However, some worldwide distributors like 'Farnell' and 'Digi-key' provide those also in small quantities (postage& package cost might be quite high..).
I personally have used Chemicon's PXA-capacitors, but I think that the Sanyo SPV-series (used e.g. in the XFX-cards version 2) are also as-good-as PXA-capacitors. There are several capacitors listed in the reference that can be used ("10SVP560M" [12]). E.g. models "6SVP820M" and "4SVP1200M" should be fine. It should be noted very small ESR-value or different capacitance value might have negative impact to the stability of voltage regulator. I will later make electrical simulation to find out safe ranges for the values. This aspect is briefly considered below where the operation of the switching regulator is explained.
Hints & warnings
----------------
- Take care off ESD-protection i.e. use an antistatic wrap
- Do not use too much heat when desoldering and soldering. Excess heat may be harmfull for some components. Also try to work quite fast i.e. it is better to solder few times at short periods that heating the components for one long period.
- Avoid mechanical stress. When removing old components use solder sucking tool or the solder wick so you do not need to bend components
- Try to keep capacitors clean i.e. touching capacitors with oily finger tips (use rubber gloves if possible)
- DO NOT TOUCH capacitors cases or other components with a hot soldering iron
Modification
------------
1. Measure resistance between TP10-TP9 (optional but recommended, just for you can check any shorts (= much lower resistance) after modification)
2. Desolder capacitors c136 and 143 (green ones, having labels e.g "4 3 470 6E") using some solder sucking tool. I you find it hard to remove capacitors, You can GENTLY (!!) lift old capacitor with a little screwdriver etc. in the same time as heating leads. Do not throw these away, but please send me after comprehensive tests.
3. Remove old solder and any dirt under capacitors
4. Solder small amount of solder to solder pads beforehand (to help solder surface mount components)
5. Carefully solder capacitors
- Be sure to have correct polarity. There is usually a clear polarity marking (=colored line/segment for minus-terminal) in alauminum electrolytic capacitors like. Sanyo CV-EX and Chemicon's PXA
- Be sure not to have so called 'cold' solders (solder not completely melted during soldering)
6. Measure resistance between TP10-TP9 again. If you get totally different value i.e. almost 0 ohms there is a problem (=usually short made by solder under capacitor) somewere. In that case DO not install card, before you have fixed the problem !! In the case of short you should remove caps again and start the work again.
Since, I have done the modification myself I do not request anymore any tasks to be done (as I asked before). But if you have performed the mod, please post how did you succeeded:
- Card model
- Situation before (how bad BS)
- What capacitor you used
- Situation after (any problems?)
3.4 Modification of Switching regulator Feedback (used to be "Gainward modification"
I changed the name of this chapter, since it was found the the feedback component values of the switching regulator vary between cards i.e. this 'modification' is most likely not invented by Gainward - just copied to make the cards working OK !
There are GF6800le-cards that differs considerably from each other is respect of 'black screen' or more precisely the so called 'BS-frequency' is varying a lot. Here, BS-frequency means the maximum memory clock frequency that can be used WITHOUT getting BS at any circumstances. A typical BS-affected card has BS-frequency of LOWER that at stock (700 MHz), which can not be tolerated. However, it was found that with some cards the BS-frequency did NOT increase very much after ANY modification while the others are capable of running MUCH higher frequency, say 850 MHz.
A detailed analysis with cards has shown that the difference of the performance can be explained with different values of components at switching regulator feedback. I have measured myself component values of two different cards MSI (BS-affected) aand GW (no-yet, BS-affected) using undirect methods (see detail at end of document). Values of the Leadtek 6800le-card were measured by 'zxcv' (he also made nice exprerients by changing values and got HUGE boost to BS-frequency, see post #382).
Below are listed measured values:
1. C1=82n, C2=15n, C3=135n (Gainward 6800le version 1b)
2. C1=40n, C2=30n, C3=120n (MSI 6800le version 1, by 'money2')
3. C1=47n, C2=33n, C3=22n (Leadtek 6800le version "4 6 470 6E, by 'zxcv')
4. C1=100n, C2=10n, C3=10n (Leadtek 6800le version "4 6 470 6E, OPTIMAL NEW VALUES !!! by 'zxcv')
The capacitor labels are related figure below
http://img148.echo.cx/img148/2742/memfb17hp.th.jpg (http://img148.echo.cx/my.php?image=memfb17hp.jpg)
Figure 21. Locations for the switching regulator feedback components (memory voltage)
The difference between values is not large, BUT effect to performance is HUGE. I made some electrical simulations to test the effect of different values. The frequency responce of the switching regulator (memory part) is shown below:
http://img259.imageshack.us/img259/1429/memfbsimu08sr.th.jpg (http://img259.imageshack.us/my.php?image=memfbsimu08sr.jpg)
Figure 22. Simulated frequency response of the switching regulator with different component values
Some electrical backround is required to see what is purpose of those curves, but 'the higher (stabile) gain is obtained in higher frequency the faster is response' -> larger BS-frequency.
It was found the Gainward has used this trick (changing values) afterwards, since I found a 'version 1b'-card:
- Some components are 'misaligned' (see figure below), like someone soldered by hand
- Some (all) components are 'older' that version 2 card (this can be verified by VERY carefully inspection of the the MOSFETs since there is a manufacturing week printed :D )
- 'OK' label has smaller value than with 'version 2'
- Capacitor 'c141' is the same (?) as in the first version
http://img205.echo.cx/img205/9743/gv6800lev1bv2botzoom32my.th.jpg (http://img205.echo.cx/my.php?image=gv6800lev1bv2botzoom32my.jpg)
Figure 23. A zoomed area of the Gainward 6800le version 1b and 2 (switching regulator part)
Exactly the same kind of cards were reported by others (e.g. 'jabjab'), so it is quite clear the card I tested was not faulty but a number of cards has been modified by GW.
This modification can be also tested by other users, who are suffering low BS-frequency. However, please read following precautions before perform this modification:
- Changing of small SMD (surface mount) capacitors is RISKY and DIFFICULT !!
- If you still want to do it, practice yourself with old/broken PCB
- Buy a set (several different values) of suitable size ceramic SMD capacitors
- Measure resistance above capacitors before starting to desolder !!! After soldering new capacitor you should have the same value (=different means short/near to short)
- Use the values above (C1=82n, C2=15n, C3=135n or C1=100n, C2=10n, C3=10n) as a starting point. Test performance and use trial-error procedure to find optimal values
If you intend to perform this mod do not hesitate to ask more instructions & hints, and please post the results !!!
3.5 Memory voltage modification
It has been reported that adjusting a memory voltage has some effect of BS [1]. There is mentioned that lowering voltage a little bit 2.72 V -> 2.68 would remove BS. And after adding a capacitor modification the memory voltage can be increased some amount without getting BS. This can be usefull for testing purposes e.g. when overclocking the memory clock frequency.
A memory voltage modification has been proposed [4].It is quite hard to perform, since there is quite high risk to destroy the card when soldering leads to surface mounted components
I found a different way to alter memory voltage either upwards or downwards as explained in figure below.
http://img216.echo.cx/img216/4646/vmemmod16ir.th.jpg (http://img216.echo.cx/my.php?image=vmemmod16ir.jpg)
Figure 24. Memory voltage modification
This method provides a very easy way to modify memory voltage and you can test it even without soldering the leads ! Please comment if this modification has been proposed earlier so I can give credits (I have not seen before).
3.6 Power supply
It has been found that power supply of the PC has (considerably effect of BS-feature. As a rule of thumb: 'the better power supply you have the smaller is propability to get BS'. Here, better power supply means stabile 5 V and 12 V output voltages i.e. how good current supplying capability there is in molex-connector for the VGA-card. In some case BS-feature has disappear (=less frequent) when having a dedicated molex-line for card i.e. not other device like hard disk connected into the same line. I demonstrated the effect of the power supply with two examples.
I had a poor quality power supply and got BS very easily e.g. with SkiAlpin. After I installed much better supply the BS-feature disappered with SkiAlpin (using Win98), but did not disappear completely.
When measuring current consumption of BS-affected MSI-card (by 'money2') I found that a VERY small series resistance in the molex-line ( 25 - 50 milliohm!!) due to measurement setup. This resistance lowers effective voltage for the card little bit and the effect was HUGE. I got BS almost immeadiate, while testing non-BS card there was not difference.
Therefore, I think that power supply effect can explain (partly) the different behaviour of cards when installed to different computers. The supplies have little bit different output voltage level and this might be enough to trigger BS-feature mouch more easily.
4. Theory of BS-feature
The theory is present here based on the investigations of Zadah, Bilabong. They have posted *a lot* of material and it is quite hard to figure out what was their final counclusion of 'what causes BS, and why?'. I wonder that the theory of 'over-voltage-protection of the Hynix-merories' is still most popular (last?) theory in the forums. Please post if I have miss this aspect !
Anaway, I think that a proper theory should not conflict following facts, starting very general ones and some related to BS-feature (I am quite sure about these, comments please!):
F1: Exactly same reference PCB is used for many 6800 and 6800LE cards (6800NU) e.g by Gainward, MSI, Leadtek ..
F2. These gards (6800 and 6800LE) are using *quite* similar components, (except GPU of course, which is modified by disabling some units for LE/NU)
F3. All IC's of the switching regulator powering the memories (and GPU) are identical ((?))
F4. Different Hynix-memories are utilized (at least 2, 2.2, 2.5, 2.8 ns), BUT there is no logic between 6800 and 6800LE memory utilization i.e. some memory model would cause BS
F5. Passive components (capacitors and inductors) vary considerably between cards
F6. ONLY some 6800(LE) models have BS-feature !!!
F7. Some cards have BS-feature immediately e.g. GW6800LE 'version 1'
F8. Some cards might work for a while, say ~ 2-3 months, before BS-feature appears e.g. GW6800LE 'version 2'. (I think that MSI6800LE belongs to this category ?)
Moreover, a real theory of BS should give decent answer to following questions:
Q1: Why some models give BS and while others are working perfectly ?
Q2: What is the difference between those cards ?
Q3: Why some models work for a while ?
Q4: Why Gainward did not get the second card working ?
Q5: Why the cards are working with the capacitor mod?
Quite tricky questions, or what you think ? Can you answer ? I have a theory that *might* not be correct one BUT it would give, I think decent, answers to those questions.
At first some talk about the operation of the used power supply for the memories, please skip if you do not like electronics..
RoHs-issues
One very important aspect, RoHs-compliance, related to capacitance deterioration was mentioned by 'Doctor6000'. The RoHs-compliance means a new standard of using non hazarous material in electronics manufacturing [18]. In practice (in the case of GF6800le) this means they have used a new solder which higher melting point that regular, say 260 C degrees. This has been 'proofed' by me and others, since the desoldering of the capacitors is more difficult that usually (=solder does not melt easily). However, the 'green' capacitors are NOT RoHs-compatible according datasheet !!! The practice would mean the capacitors might have been (partially) ruined when manufacturing of the cards. Moreover, that would give a 'clean' explanation why some SVP-capacitors (MSI and XFX-cards) have been ruined, althought having a proper ripple current value. Also that would explain easily why some manufacturers do not have problems at all: they do not use Rohs-compatible solders !!!
4.1 Switching power supply for memories and GPU.
ISL6534
The supply voltages for the memories and GPU in the 6800(le)-cards are generated with a swithing regulator ISL6534 [5]. I draw a simplified schematic for the most important parts of the swithing regulator shown in Fig. 11. Please, note that only RELEVANT components have been drawn and there CAN be errors (it is not so easy to re-engineer multilayer PCB ;) ). At least plenty of capacitor are missing: there are many ceramic (1u - 10u, I suppose) capacitors placed all over the board to keep the power supplies clean.
In short: ISL6534 generates pulses driving MOSFET's and the squarewave output is soften with the coil and capacitor(s). The pulsewidth of the pulses are adjusted with feedbacks from memory and GPU voltages. So, if a larger current is required the upper MOS is adjusted to conduct longer period (the frequency is fixed to 300 kHz). Please refer to datasheet for detailed information. Some important details of the regulator:
http://img121.exs.cx/img121/8189/fig78fq.th.jpg (http://img121.exs.cx/my.php?loc=img121&image=fig78fq.jpg)
Figure 25. A simplified schematic of the switching power supply of the 6800-cards (for memories and GPU)
- Power supplies for ISL6534 VCC, VCC12 (5v, 12V) are NOT taken from the MOLEX-connector, but they are coming from AGP
- GPU-voltage is generated from 12V line from MOLEX
- Memory-voltage is generated from 5V line from MOLEX (!!)
I think that the following detail is the most important when considering BS-feature:
!!! ISL6534 DO NOT HAVE HAVE OVER CURRENT SENSING CIRCUIT AT ALL !!!
,and they are using diffenent methods to protect the IC from short circuit - and I think that is a key point of BS-feature. The short-circuit protection is explained as follows [5]
"There is no current sensing or rDS(ON) sensing or Under-
Voltage sensing on the ISL6534. However, if either Channel
1 or 2 output is shorted while active, there is a simple
detection on the error amp COMP output that implies either
Over-Current or Under-Voltage; the PGOOD pin goes low
immediately. If the condition persists for 1-2 internal clock
cycles (3-6µs at 300kHz), then ALL 3 Outputs are latched
off, requiring either a VCC or VCC12 POR to restart. The
protection was not designed to work for the case of powering
up an output into a short-circuit, and there are limitations on
detecting applied shorts."
In other words 'a short' is detected if the output voltage cannot be raised to specified level during 1-2 cycles. If 'short' is detected all outputs are switched OFF i.e. shutting down the GPU and memories. This will be considerd in detail in next chapter.
APL5331
DDR-memories require a 'VTT' and 'VREF' signals as reference levels for the input and output. These nodes either sink or source current and therefore a special regulator is required. ISL6534 can be connected in such manner that it generates those voltages, but these cards utilizes linear regulator APL5331 for that purpose [6]. This is somehow important aspect since this regulator increases the loading of the memory voltage.
HYNIX-memories
I think that all 6800(LE) cards are using Hynix-memories: HY5DU283222AF-xx [7]. There are some variations about speed, some cards have AF-22 and the others AF-28 etc. but basically they are quite the same chips (?). There is a single datasheet that specifies absolute maximum rating etc. so I think that they measure the speed of the chip and sell it 2.2 ns or 2.8 ns memory depending the results (such like CPU's).
I wonder that there is NOT any overprotection circuitry within those memories, in spite of the some posts. Or at least when having voltage levels << 3 V (the absolute maximum ratings for VDD is 3.6 V !!). Moreover, the termination voltage of those memories means that the signals are TERMINATED to that voltage VDD/2, so it does NOT mean 'chip will shut down when abnormal VTT voltage is detected'. Anybody expert with memories ? Please comment !!
Feedback-loop
After investigating GW version 1b -card (which had a manufacturing fault) I found how critical it is in respect of the stability of the switching regulator. In addition to ESR of the output capaciotrs (c136 and c143) at least following components have major effect to the performance of the regulator:
- All output capacitors for memory voltage and their ESR and ESL
- Inductor
- Capacitor and resistor values at the feedback loop (these were altered for GW version 1b..)
- Load resistance !! i.e. current consumption of the memories
--
Before I found the information about Sanyo CV-EX-series capacitors, I tried to make specifications on my own.
Using the equation in the datasheet of the switching regulator [5] it can be calculated that the ESR-value of the capacitor has to match to capacitance and inductor values of the output filter. This gives relationship of
ESR < SQRT(L/C) (ESR = Effective series resistance, L = inductance, C = capacitance)
Having L = 1uH ja C = 2*470 uF = 940 uF, gives ESR < 0 .033 ohm or ESR < 0.066 ohm/each capacitor (two in parallel). This value matches quite well to datasheet value.
--
The ripple current rating is a little bit tricky. The measured value for current consumption is between 0.6 - 2.64 A for 5V molex-connector which corresponds to power of 13.2 W. Power power can be much more higher due to the limitation of the multimeter current measuring badtwidth. If it is assumed that the power taken from molex-connector in this case is mainly consumed by memories and the efficiency of the switching power is 80 %, the maximum power consumed by memories is 0.8*13.2 W = 10.56 W. This corresponds current of ~3.9A @ 2.72V (memory voltage). The maximum power consumption of single Hynix-memory is 2W, so 8*2W is 16 W for all memories. Therefore the calculated value is at least in the same magnitude.
Peak current and voltage of the switching power supply can be approximated as [5]
I(ripple) = ((Vin-Vout)/(Fs*L))(Vout/Vin) (Vin = 5V, Vout = 2.72 V, L = 1uH, Fs = 300 kHz)
Vout(ripple) = I(ripple)*ESR (ESR = 0.030 ohm per each capacitor WHEN NEW ??)
This gives an approximation of I(ripple) ~ 4.2 A . There are two capacitors (C136 and C143) in parallel at output of the switching regulator, so each capacitor has to handle current of ~ 2.1 A When comparing this value to the specifications of Sanyo CV-EX capacitors (470 uF, 6.3V, I(Ripple) = 2.09 A !!!) it can be easily seen that the ripple current rating should be proper fo this purpose. But as explained above the approximation is more likely too low so the current rating is too tight is in normal condition. In addition to the ripple current the capacitors are charged/discharged heavily as explained above (consumption measured for molex-connector changes between 0.6 - 2.64 A). Moreover, the swithing regulator is oscillating (see 5.4), which causes VERY high varying current at frequency range between 100 Hz - 1000 kHz. I have not managed so far to estimate current value due to oscillation, but while having noisy audio at the CD-rom output the current level HAS to be remarkable (I will later try to measure it). Aluminium electrolytic capacitors are NOT designed for charging and discharging frequently. These aspects might be the key elements for the reason WHY capacitors fail as a new or in same case can withstand for a some time. The excess ripple current and/or high charge/disharge of aluminium electrolytic capacitors causes [9][10][15]. In addition to ripple current effects the likely utilization of 'RoHs-compatible' solder has a considerably effect of the capacitor life i.e. the capacitors will get old/ruined instanteously during manufacturing.
- Life time decrease
- Capacitance value decrease
- Increase of ESR-value
.. which all can be easily cause power supply to fail in hard load condition -> BS.
--
I have tried to simulate operation of the switching power supply. I have used a free simulator iSim can be downloaded here [17]. The schematic example for switching regulator can be also download from there after registration. I use component values obtained methods explained Ch. 5.6. It should be noted these values are not very accurate and the obtained results gives only insight about what is happening. Moreover, a number of parasitic components (resistances, inductances, capacitances) due to PCB etc. are not included to the simulation and these could have relatively large impact to overall performance.
I have been simulating AC-response and transient for the feedback loop of the memory switching regulator. I have some preliminary indicating, that the memory part of the regulator is vulnerable to unstabile operation !! I will post result later when a more accurate values for e.g. current consumption for molex-connector is available.
It is kindly asked to somebody measure feedback components (see Ch 5.5) for BS-affected card (different to MSI), so we (I) could verify weather the component values have been altered for Gainward version 1b card.
4.2 Black Screen
Here we are..what causes black screen ? My engineering guess in one word - capacitors. I think that following happens to make the BS appear. At a starting point there is a moderate load for the graphics card (this means e.g. when you launch the 'SkiAlpin' and in menu you start the game). Suddenly, the program start to write (or read) gigantic amount of data to the memory from GPU (you start skiing in 'SkiAlpin'). The memory chips sink as much current as they can for some clock cycles. This current is taken from the capacitors 'C136' and 'C143', since the switching regulator is running at frequency of 300 kHz - memories 350 MHz (700DDR). IF there are good capacitors having a low ESR (effective series resistance) and capacitance value high enough - no problem there is plenty of charge reserved and switching regulator will start pump more current within next pulse. But if there is *something bad* with those capacitor following might happen:
1. Memory voltage drops lower level
2. ISL6534 detects 'short'
3. ISL6534 tries to increase pulse width
4. If short is not disappeared within 1-3 (?) pulse all outputs are shut down !
5. Memories and GPU lose power
6. Black screen
There is an interesting comment in the datasheet of ISL6534 [5]:
"In general, the faster the rise time of the output current
during the short, the more current will be allowed on the
initial peak, and the better chance the COMP pin will have
a sharp rise as well. A low resistance short (#4) and a
higher output voltage (#5) both help. However, if the
current ramps too fast, then a false trip is also possible
(shutting down at a current level still within the expected
load range)"
.. so when having a fast current peak (=voltage level drops rapidly) is likely that power will shut off. After the BS ISL6534 is still running (see measurement results), but all outputs are 'off', which 'defends' the theory.
Another point of view: if the BS is causes by *over voltage*, how a over voltage is generated ? I think that only way to have a overvoltage is to increase the pulsewidth for the ISL6534, but what would be the mechanism to generate such situation ?
Moreover, if the memories were shut down BEFORE ISL6534, I think that the memory voltage should remain 'up' after shut down since there is not specified a lowest current that ISL6534 can drive (the minimum pulsewidth is ~0).When debugging this problem I thought that ISL6534 could be shut down due to the poor molex-lines i.e. the power supply voltage for ISL6534 would be too noisy. Therefore I measured VCC and VCC12 and found that those voltages are NOT taken from molex-connector, but AGP-bus ! Therefore noisy molex-line (due to switching power itself) cannot cause the shutdown of ISL6534.
So, I think (=I am sure) that BS is due to ISL6534 is shutting down the outputs due to faulty detected short. Yes, I know what you are going to ask: why some cards are working ? Well.. I have been investigated *a lot* the figures of the different cards and I found that the memory capacitors of these cards are the same - or...? As stated above I found that there are three numbers printed with green on the top of those capacitors like "4 3 470 6E". The middle value is capacitance in microfarads and '6E' stands for 6.3 V (I think so), but .. the first row has to means a lot number. This is not 100 % sure since I have not found the manufacturer (and datasheet) for those capacitors, but usually there is a lot number with that kind of capacitors. I found that BS-cards have certain lot numbers e.g. GW6800LE V1 "4 3". And more interesting thing two BS-cards from different manufactures have the same number like GW6800LE V2 and MSI6800LE "4 8".
Moreover, I have not seen any 6800(LE) other card (than BS-featured) with that lot number. Has there been a 'little' problem with those lots ??? I believe, yes. A lot means a large batch of components etc. manufactured in the same time, using the same materials, manufactured in the same factory etc. , so different lots can be totally different ! This application is very sensitive to following parameters:
- ESR (effective series resistance) since it determines the maximum available peak current
- Capacitance value (=how long capacitor can provide current, before voltage level drops certain amount)
- Response of the feedback loop (see Ch 3.4 above)
These kind of capacitors (aluminium electrolytic capacitor) are not so easy to manufacture and they are vulnerable to many aspect like: soldering, ripple current, temperature etc. Therefore, it is not unlikely that when buing such capacitors as-cheap-as-possible you can get very bad capacitors. I suppose that these lots "4 3" and "4 8" get somehow ruined - but it was not detected (or informed!!!). The first lot was totally bad since the cards having those capacitors made BS-immediately. Afterwards, manufactured (=Sanyo) fixed something with "4 8" and the cards might work even 3 months ! What can be wrong with those capacitors ? I think the initial capacitor values has to be correct - that would certainly be detected. Large ESR-value is harder to measure, and therefore that could be the reason the first GW-gards failed: capacitor are not capable to provide required current. The second lot "4 8" usually is working at first, so maybe they found the high ESR-value and did something;). However, the quality of the capacitors is so low that that something happens again when time goes on. A nice diagram about aluminium electrolytic capacitor failure mechanism can be found here [9]. As you can see almost anything can be happened e.g. if the ripple-current exceeds the nominal value.
A very interesting aspect about temperature effect was reported in this same thread (see posts #3, #6, #7, Thank you ZeroHero !). It was found that by cooling the ambient of the card ~ 10 degrees causes immediately BS under suitable conditions. This phenomena can be caused by many factors e.g. cooling of the switching transistors. However, the on-resistance of the transistors reduces when temperature is lower and the operation of the switching power should be better ! On the other hand the ESR-value of the aluminium electrolytic capacitors INCREASES when temperature REDUCES, which LOWERS the performance of the switching power supply (See another document about the aluminium electrolytic capacitors if you are very interested [10].). In order to make this even more complex the higher resistance (ESR) causes higher voltage drop over capacitor and in the same time the power 'burn' in the resistor increases, which increases the temperature (would be nice to make electrical model for this :D ).
This theory might be possible to proof by measuring those capacitor very accurately. This would require removing a capacitor from PCB and using a high performance inpedance meter. I am voluteer to make those measurements if someone has totally burn card and he/she is willing to provide those capacitors (either BS or non-BS cards). Please post or send a private message, so we could think how to arrange this. This would also be useful in sense of finding the final failure mechanism of those cards: will it be a short (very bad !!!) or open. In the version 2.2 this has been proofed (=measured high value of ESR), see the and of the document for measurement results.
4.3 Answers for those questions
Using proposed theory *some kind of * answers can be given:
Q1,Q2: They have poor capacitors for another manufacturing lot !
Q3,Q4: Those capacitors get *old* very quickly - some parameters (ESR or capacitance) deteriorate by temperature/ripple-current etc. But when they are new no-one cannot find anything bad !
Q5: Adding a parallel capacitor, 'overrides' the bad one - it provides enough current when high ripple current is drawn by memories
5. Measurement results
I measured several parameters from the unmodified 'GW6800le version 2'. After I got oppotunity to measure version 1B card and I performed all measurements again (!) I list below all results, some of these are not so relevant but maybe someone can use those values e.g. for debugging some other problem (?). The locations of the testpoints are shown in Fig. 13, which make it more easy to locate them when measuring a running PC. Please note the locations for TP5 and TP6 ! Those TP's were changed in the first versions of this document (1.0 - 1.6), but now tables and figure below should be ok..
http://img148.echo.cx/img148/3621/6800tps28bf.th.jpg (http://img148.echo.cx/my.php?image=6800tps28bf.jpg)
Figure 26. Test point and "PHASE"-node locations in the 6800(le)-card (bottom side of PCB).
5.1 DC-current
I measured DC-current consumption of the 5V and 12 V Molex-lines for the card. The lines provides most of the power for the card, so measuring these gives a good approximate about loading state of the card. It should be noted that these measurements are not very accurate, since the current level was changing rapidly. Therefore I provide a maximum value that I could read.
DC-current [A], vsync = off
'Version 2'
case | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
-------------------------------------------------------------------------
5V | 0.60 | 0.79 | 1.68 | 2.28 | 0.97 | 1.90 | 2.64 | 2.36 |
12V | 0.79 | 1.13 | 1.33 | 1.30 | 1.18 | 1.29 | 1.60 | 1.52 |
Please note that these values are BAD due to manufacturing bug !! (see text) 'Version 1B" (higher values due to higher voltage levels!!)
case | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
-------------------------------------------------------------------------
5V | 0.94 | 1.18 | 1.50 | 2.50 | 1.90 | 2.50 | 2.89 | 3.23 |
12V | 0.89 | 1.26 | 1.33 | 1.44 | 1.50 | 1.45 | 1.68 | 1.74 |
1. Boot
2. Idle (win98)
3. SkiAlpin -demo (main menu, win98)
4. SkiAlpin -demo (in game, win98) -> no BS
5. SkiAlpin -demo (main menu, win2000)
6. SkiAlpin -demo (in game, win2000) -> BS (for version 2)
7. 3dmark03 (Game test 1)
8. AquaMark03
It can be seen that BS occured at moderate or not maximum power level (only with version 2!), which indicates the power fails rapid increase of current not heavy but stable load.
5.2 DC-voltage
I measured Voltage from TP1 - TP14 and the 5V and 12 V Molex-lines for the card before and after BS-occured.
'Version 2'
point | TP1 | TP2 | TP3 | TP4 | TP5 | TP6 | TP7 | TP8 | TP9 | TP10 | TP11 | TP12 | TP13 | TP14 | 5V | 12V |
no-BS | 2.72 | 0.00 | 1.37 | 0.00 | 0.00 | 1.36 | 3.35 | 0.60 | 0.00 | 2.72 | 1.12 | 0.61 | 3.31 | 2.72 | 5.04 | 11.89 |
bS | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.26 | 0.05 | 0.00 | 0.00 | 0.00 | 0.00 | 3.32 | 2.72 | 5.04 | 11.89 |
Please note that these values are BAD due to manufacturing bug !! (see text) 'Version 1B'
no-BS | 3.19 | 0.00 | 1.61 | 0.00 | 0.00 | 1.60 | 3.39 | 0.61 | 0.00 | 3.18 | 1.11 | 0.60 | 3.31 | 2.72 | 5.04 | 11.93 |
TP1=TP10 (memory voltage)
TP11 (GPU voltage, small resistor 1 ohm)
TP3, TP5 (VTT/VREF voltages)
It can be clearly seen that ISL6534 operate after BS, since some reference output are working, but all switch regulator outputs have been shut down.
5.2 Resistance and capacitance
I measured the resistance and capacitance between listed nodes and GND. Please, note that capacitance values starting with ~ means uncertain value (oscillation) and the accuracy of the some capacitance values is poor (multimeter cannot measure capacitance value correctly, when there is a small parallel resistor).
'Version 2'
point | TP1 | TP2 | TP3 | TP4 | TP5 | TP6 | TP7 | TP8 | TP9 | TP10 | TP11 | TP12 | TP13 | TP14 | 5V | 12V |
resistance | 101.3 | 0 | 161.3 | 0 | 0 | 527 | 852 | 851 | 0 | 101.2 | 5.6 | 545 | 5.93k | 4.86k | 4.8k | 7.8k |
capacitance | ~5.7u | 0 | 1.46u | 0 | 0 | 125n | 20.9u | 17u | 0 | 7.9u | ~150n | 11n | 87.7n | 96.5n | 710u | 780u |
'Version 1B'
point | TP1 | TP2 | TP3 | TP4 | TP5 | TP6 | TP7 | TP8 | TP9 | TP10 | TP11 | TP12 | TP13 | TP14 | 5V | 12V |
resistance | 70.5 | 0 | 111.3 | 0 | 0 | 520 | 851 | 849 | 0 | 70.6 | 3.6 | 543 | 5.94k | 4.87k | 4.65k | 7.8k |
capacitance | ~4.7u | 0 | ~0.5u | 0 | 0 | 129n | 18.7u | 15u | 0 | ~4.7u | ~100n | 7.5n | 103nn | 111n | 702u | 772u |
5.3 AC-voltage
I measured the 'phase' node and memory voltages (see page 8 [4]) of the ISL6534 regulator circuit with HP 54602A oscilloscope. Yes, it was quite tricky operation but I succeeded ! I tried to measure if the voltage level would behave abnormally before/after BS. I found that the used oscilloscope was not able to show anything important when measuring memory voltage - the accuracy (speed) was not sufficient to detect BS. In theory that would be possible to measure, but this measurement requires a *high* cost equipment. Anyway, I managed to measure phase signal and found nice 299 kHz square wave signal having amplitude between 0 - ~5 volts. The rise and the fall times of thes signal were 13ns and 3 ns, respectively. The higher risetime is due to lower gate drive for upper MOSFET (see datasheet if you are really interested [5]). The pulse width (PW) varied rapidly when a heavy load was applied. The pulse width was 54 % in the idle state and I saw values > 60 % under heavy load. However, I did not saw pulse width higher that those values when BS occured. Again, this would indicate that BS happens at medium load by very rapid peak current that cannot ((easily)) be measured.
I also measured AC-voltage of the TP10 (Vmem) with a high performance instrumentation amplifier constructed with a PGA202. This circuit provides a high accuracy adjustable gain and rejects all common mode noise, so it is ideal to measure ripple voltages. I connected measurement probe between TP10 and TP9 as shown in foto below.
http://img405.imageshack.us/img405/2267/vmeassetup01qb.th.jpg (http://img405.imageshack.us/my.php?image=vmeassetup01qb.jpg)
Figure 27. Memory ripple voltage measurement setup
The gain was adjusted G=4, so the measured values has to be divided by four to get actual ripple voltages. Below are shown two measured cases: (a) windows in idle state (b) under heavy 3d-load. It is clearly seen that ripple voltage increases in order of four under heavy load, which indicates a high ripple current (see below).
http://img405.imageshack.us/img405/8997/vripple16mp.th.jpg (http://img405.imageshack.us/my.php?image=vripple16mp.jpg)
http://img405.imageshack.us/img405/5859/vripple20bw.th.jpg (http://img405.imageshack.us/my.php?image=vripple20bw.jpg)
Figure 28. Measurered memory ripple voltage (a) windows idle (b) under heavy 3d-load
5.4 AC-current
I measured the the AC-current consumption of the molex 5V line (which provided current for the memory regulator) as follows. I inserted a small ( 50 milliohm) in series with 5 V line of the molex-connector. Above the resistor I put a instrumentation amplifier having gain of 20 to get a reading of 1 A = 1 V in the measurement device. The foto of the setup is shown below.
http://img368.imageshack.us/img368/6728/imeassetup05am.th.jpg (http://img368.imageshack.us/my.php?image=imeassetup05am.jpg)
Figure 29. Molex 5V line ripple current measurement setup
I used an oscilloscope in different conditions to see how current consumption is varying. In the first foto below (a) is shown a windows idle state, and current spikes at low frequency is detected. This condition proofes the origin of noisy audio, since the frequency lies on audio band and traces the one that can be heard !! The rest two cases show how 'ugly' the current consumption variation can be. The current consumption varies between 1 A - 4 A which is quite near the values approximated above.
http://img455.imageshack.us/img455/2995/iripple35pn.th.jpg (http://img455.imageshack.us/my.php?image=iripple35pn.jpg)
http://img294.imageshack.us/img294/8066/iripple19ah.th.jpg (http://img294.imageshack.us/my.php?image=iripple19ah.jpg)
http://img369.imageshack.us/img369/9012/iripple24hf.th.jpg (http://img369.imageshack.us/my.php?image=iripple24hf.jpg)
Figure 30. Measurered molex 5V line ripple current (a) windows idle (b)(c) under heavy 3d-load
5.6 Noisy audio measurements
I have been solving almost whole the time when using 6800le-card quite annoying audio/noise problems. This problem has been very hard to solve, and took long time to solve it or find a sensible reason. Now I am sure that problem is caused ONLY due to GW6800le-card, and therefore I will explain the 'feature' in this document. It is likely, that the same problem disturbs also other users having a BS-card and I have demostrated excatly the same behaviour with a card never (yet) shown BS-feature.
The problem is simply a annoying sound or noise, having varing frequency and amplitude. Moreover, the frequency is 'modulated' with the current framerate (!!). Usually the noise has quite low amplitude and it can be only heard when there are no other sounds. For instance a 3dmark03-benchmark is good for testing purposes, since it does not have (mostly part) sound and the framerate is varing considerably. Following setup is required for the audio test:
- Computer having a GF6800(le)-card, that is known/assumed to be BS-affected
- CD-rom having audio cable installed to soundcard/mainbourd (important!!)
- In the 'Volume control'-panel the 'CD-player Volume' is maximized
- A suitable program is launched e.g. 3dmark03 or aquamark03 (without sound)
- Headphones preferred, since the amplitude of the noise might be low
- Just listen ..
Why I am sure it is caused by GF6800 ? I have tested a lot following different options/combinations:
- GF4200 : No noise
- GW6800le version 2 vs version 1b : No difference !!! version 1b has not made BS (yet) !!!
- Win98 vs Win2000 : No difference
- Sounblaster Audigy vs integrated : No difference
- Different CD-players : Minor difference (DVD-player makes louder noise, since it consumes more current)
- Molex removed from CD : No noise
- Added 'filter' to CD-molex : Very low noise
The filter I used is constructed with wires ripped from old PC-power and a large ferrite ring (ferrite rings can also be 'found' from old electronic equipments e.g. power supplies). The molex-wires have been cut in the middle and wounded several times around the ferrite. Finally wires are soldered together again, see figure below.
http://img245.echo.cx/img245/7923/molexferrite1oi.th.jpg (http://img245.echo.cx/my.php?image=molexferrite1oi.jpg)
Figure 31. A lowpass filter for removing power supply noise due to GF6800-card
In order to have some idea what kind of sound you would except, I recorded following audio tracks (amplified a lot): 'Windows 2000 - do nothing', ' 3dmark03 GT1', 'Aquamark03' and 'SkiAlpin'. If interested, download and enjoy :D
http://www.angelfire.com/electronic2/sikaripp/6800le/win2000_idle.mp3
http://www.angelfire.com/electronic2/sikaripp/6800le/audio_3dmark03_1.mp3
http://www.angelfire.com/electronic2/sikaripp/6800le/audio_aqua03_1.mp3
http://www.angelfire.com/electronic2/sikaripp/6800le/audio_skiapline_1.mp3
Why this is happening ??? And why it is happening also for card never made a BS-feature ? Does this noise predict BS-feature in future (if not yet affected) ?!?! I have a theory and I will explain it in more detail in the next release. In short (for this release) I think that switching regulator for memories is OSCILLATING at quite low frequency and this adds noise to the molex lines (mainly 5V).The noise is coupled via CD-player to the audio cable, due to poor power supply rejection ratio at low frequencies.
I performed more audio measurements having setup as follows. I utilized case 'windows 2000 is doing anything', to have a constant current consumption i.e. oscillation at fixed frequency. I recorded noise audio in four different cases:
- No filter at molex-connector
- 12 V molex-line filtered (12 V wire wounded around ferrite)
- 5 V molex-line filtered (5 V wire wounded around ferrite)
- Both 12 V and 5 V molex-lines filtered
Afterwards I concated files together and amplified data to make 'noise' audiable. When listening constructed file (link below), it can be easily noticed the difference due to filtering options. The filtering of 5V-molex line has much bigger effect than the 12-V line, so the main source for noise is 5V-molex line. Since, this voltage is used for memory voltage generation (and 12 V for GPU voltage), I suggest oscillation of switching regulator as a mechanism for noisy audio generation.
http://www.angelfire.com/electronic2/sikaripp/6800le/win2000_idle_all_2.mp3
5.5 Temperature measurements
I have measured temperatures of the cards: 'Gainward 6800le version 1b' and MSI 6800le (BS-affected). I used a Pt100-sensor (high accuracy platinum temperature sensor) glued at top to wooden stick (see foto below), which I gently put top of each component.
Figure 33. Temperature sensor (Pt100) attached at the tip of wooden stick
The accuracy of the sensor is high, but due to poor thermal contact the measured temperatures are too optimistic (=one or two degrees too low). Results for both card can be found below:
http://img442.imageshack.us/img442/8919/gwv1bowntemp28kt.th.jpg (http://img442.imageshack.us/my.php?image=gwv1bowntemp28kt.jpg)
http://img401.imageshack.us/img401/7384/msiv1m2temp27nx.th.jpg (http://img401.imageshack.us/my.php?image=msiv1m2temp27nx.jpg)
Figure 34. Measured temperatures of GW6800le and MSI6800le-cards
It can be seen, that the temperature of the 'green' capacitors is considerably higher in BS-affected card (MSI), which makes sense since higher ESR of capacitor means higher power consumption and higher temperature.
I measured or tested effect of very low temperature to 'Gainward 6800le version 1b' -card. It should be noted that this card has never (yet) made 'black screen', and therefore it is kindly requested wheather somebody could repeat this test with BS-affected card. Detailed instructions can be found in 'Most wanted'-list at the end of this document.
I used a freezing spray especially designed for cooling/testing of electronic equipment. If somebody will do the test it is important to use that kind of freezing method to avoid shorts etc. I launched 'the best BS-program I know' SkiAlpine2005-demo having cold spray in hand. When applying a hard 3D-load for card (=skiing down the route) I simultaneously freezed those capacitors (C136 and C143). I used cold spray A LOT to make those capacitors fully icy i.e. the temparature became well below 0 C (celcius). I did not get BS even trying this trick for little period (I wonder I am the FIRST person in this planet that have desperately tried toget BS to non-BS-card..).
Another interesting thing was that the frequency of the noisy audio did not change during freeze.
This test showed that when new (=having still good capacitors) these cards are not very sensitive to temperature. In fact in the datasheet of those capacitors is stated that they should have excellent temperature characteristics (when compared to standard aluminium electrolytic capacitors). But what kind of temperature sensitivity those capacitors have when they are almost worn out (dead) ?
I tested also the 'temperature trick' for the BS-affected MSI-card. I was running a program that did not give before BS (AquaMark03) and freezed capacitors with the spray: the result was a sudden BS !!! So, this test proofed (once again) that the problem is ESR, since ESR of these capacitors is increased when temperature goes down.
5.6 Component value determination
I have been investigated and measured the components that are part of the switching regulator. The aim of this work is to understand what the high ESR value of the capacitors causes and also find out what Gainward has been modified in the feedback path for version 1b. I think that I have now values for all relevant components and know the circuit connections at required accuracy to sketch the switching power circuitry for memory voltage generation. In this point I have now considered core-voltage generation, since I do not think it is so important for this problem.
The schematic for the switching power is quite similar to shown in the datasheet of the ISL6534 (page 17). The most infortant components in respect of the transient characteristics of the power supply are:
- Mosfets
- Inductor (output filter)
- Capacitors C136 and C143 (output filter)
- Compensation circuit
The aim of the compensation circuit is to provide proper phase margin for signal i.e. prevent oscillations etc. when heavy load transients are applied. I have been trying to analyze and make electrical simulations about this cicuit, but in order to this exact component values are required. Below I decribe in very detailed a method how I measured the component values of the feedback path. In order to find out the differences between cards (=explanations why some card are working), it is asked wheather somebody could make these measurements. These measurement requires a multimeter having capacitance and resistance measurement capability. If you will perform these DO NOT measure while power is 'ON' and be carefull not to destroy your card due to ESD (static electricity discharge)!
Following instruction are related to names shown in the Figure below.
http://img148.echo.cx/img148/2742/memfb17hp.th.jpg (http://img148.echo.cx/my.php?image=memfb17hp.jpg)
Figure 33. Locations for the switching regulator feedback components (memory voltage)
Resistors
These are quite easy to measure: no tricks are required and the polarity of the measurement does not matter (+/- -terminals of the multimeter). It is asked to measure resistance between following nodes:
R(N1-VDD_MEMORY) = 10 ohm (=R3)
R(N2-N1) = ~ 1000 ohm (=R1)
R(N2-N3) = ~ 6850 ohm (=R2)
R(TP14-GND) = ~ 4870 ohm (=R6)
R(TP14-TP13) = ~ 1000 ohm (=R5)
R(VDD_MEMORY-GND) = ~ 70 ohm
Capacitors
This is *little* bit tricky. If you are aware of basic electronics you know that in general case a resistor or capacitor connected as a part of circuit is impossible to measure. In this case those resistors can be determined by the measurement results above . The main problem with the capacitors became to fact that HAVING A RESISTOR PARALLEL TO CAPACITOR DETERIORATES RESULT. I was thing this problem *little* and I though for a long that only way to measure those values would require desoldering of components. However, I discovered two tricks and now I am quite sure that values can be determined decent accuracy with simple measurements.
Trick '1' is to short e.g with little wires some nodes to cancel some capacitors and resistors (Warning again, never do this when power is on, I HARDLY suggest to do this when card is removed from PC), . As a results only a few components are effective for measurements and values can be more easily interprepted.
Trick '2' is to afterward alter/distort/correct measured capacitance values WHEN a parallel resistance exists. I did the trick as follows. I measured capacitance and resistance values between certain nodes. Then I adjusted potentiometer (variable resistor) to measured value add soldered little arbitrary value (!) capacitors parallel to this resitor. Using trial and error I found a capacitance value which gave exactly the same capacitance reading. When removing the resistor the real capacitance value can be easily measured ! At first I tried to analytically determine the effect of the parallel resistor but turned out very nonlinear phenomena, and I decided to use trial and error method.
The capacitance values should be EXACTLY the same using this method, but I found that accuracy is lower when having low parallel resistance i.e. the obtained capacitance value is very sensitive to resistance.
I measured capacitance in following cases to determine values for relevant components (Please note that the first node was allways connected to positive terminal of multimeter!):
Easy ones:
C(TP13-GND) = 108 nF
C(TP14-GND) = 117 nF
C(N4-GND) = 155 nF
Tricky ones (need to short some points e.g. little wires NO need to solder just push gently between those nodes):
C(N3-GND = 62.6 nF (both VDD_MEMORY and N2 shorted to GND)
C(N3&N4-GND) = 1.8 nF (here N3 and N4 were shorted with measurement cable AND both VDD_MEMORY and N2 shorted to GND in the same time... yes that one is *really* tricky)
C(N2-GND) = 35 nF (here N4 AND VDD_MEMORY AND N2 were shorted to GND)
Using this data I determined capacitances (see Figure above for labels) and rounded those to standard values:
C1 = 100 nF
C2 = 10 nF
C3 = 100 nF (+/- ~50 nF)
Please note that these capacitance values might not very accurate due to reasons stated above. However, I think that these are accurate enough to provide information about feedback loop dynamics (simulations are under work..)
5.7 ESR measurements
Finally, for the version 2.2, I finised the ESR measurementS for the C136 & C143-capacitors. It was a HARD project and took ~ half a year to finish !! I would kindly thank all how send me old 'green' CV-EX capacitors, so I got enough 'material' for measurements. In the beginning I show the acquired results of the ESR-measurements. Then I briefly explain the mesurement device and try to approximate the errors due to setup.
ESR-values
I measured the ESR-values (equivalent series resistance) of several C136 & C143 capacitors ripped off the black screen affected cards. Those capacitors are CV-EX series from Sanyo and the specified ESR should be lower than 30 mohm, 470 uF 6.3 V (milliohm). The main reason I was EXTREMELY interested to acquire actual values is that a value of 2 times higher than nominal is considerd as destroyed component. So, this mesurement should confirm (and it in fact CONFIRMED) the problem is really releted to capacitors out of specifications. (On the other hand the successfully performed ultimate capacitor modification confirmed that also before, but I WANT to see on my own eyes WHAT are ESR values).
Here are the measured ESR-values for "CV-EX, 470 uF 6.3 V"-capacitors ripped off from GW and MSI 6800le-cards:
| 1 | 1 | 1 | 1 | 2 | 2 | 3 |
---------------------------------------------------------
ESR [mohm] | 207 | 194 | 199 | 211 | 192 | 170 | 227 |
Capacitor codes:
1 = "4 3 470 6E"
2 = "4 8 470 6E"
3 = "4 6 470 6E"
I am sure that everyone ever measured ESR-values within milliohm accuracy, would be interested about measurement accuracy. I considered that issue below, but to verify the proposed measurement setup I characterized also known very high quality capacitors. Below are shown measured values of brand new Chemi-Con PXA/PXH-series capacitors.
| 1 | 2 | 3 | 4 | 5 |
--------------------------------------------
ESR [mohm] | 34 | 30 | 73 | 61 | 69 |
Capacitor codes, maximum specified ESR-value in parenthesis:
1 = "PXA, 1200 uF, 4V" (10 mohm)
2 = "PXA, 820 uF, 4V" (10 mohm)
3 = "PXA, 470 uF, 6.3V" (20 mohm)
4 = "PXA, 470 uF, 4V" (20 mohm)
5 = "PXH, 680 uF, 4V" (25 mohm)
Following observations can be made using this data:
- CV-EX capacitors HAVE BEEN RUINED, since ESR-value is many times larger than specified !!
- Measurement accuracy gets worse at very low ESR-values (see below)
- Mesurement results are monotonous i.e. higher series resitance gives higher value
- Reproductability of measurements is very high i.e. the same device gives the same value when measured again
- High accurate COMPARISON between components is possible to perform
Therefore, I propose the presented mesurement results show the measured CV-EX capacitors have MUCH larger ESR than specified than brand new PXA-capacitors, having specifiactions in the same magnitude of CV-EX-capacitors. I confirmed this by connecting a very high quality capacitor in series with ~ 100 mohm resistor and got a value of ~ 100 mohm larger than without series resistance. This experiment shows that even if the meter is not operating in a linear/high accuracy operation at level of ~ 0 ohms ESR, it can operate nicely in larger values.
ESR-meter
In very beginning I considered the hardest task would be obtaining proper amount of old CV-EX capacitors. However, because of helpfull people all over the world (Poland, Germany, Finland to name some..) I got a good set of capacitors. Later I found HOW hard it is to build a ESM-meter with milliohm (or 10 milliohm) range... I build two (or in fact three) different meter before I was happy with results.
The ESR-meter I build is based on clever and simply design proposed by Ray Porter [19]. Unfortunately the link to page that explained the operation does not exist anymore, but I'll briefly explain the idea. I am not going to describe detailed the operation and how-to-build such device, because I suppose that a person how NEEDS milliohm range ERS-meter do know something about electronics.
Operation
The simplified schematic of the ESR-meter is shown below. In the original design there was a 3-pole 4-way switch selecting feedback resistors to change operation between different ranges (1, 10, 100 ohm). If you are REALLY interested about details of that please send a private message. The idea is based on a negative impedance oscillator made with an operational amplifier (X2, in Figure). When the positive and negative (resistive) feedbacks are exatly the same the pulses generated by OPAMP X1 keep X2 in a continuous oscillation. The device-under-test (=capacitor) is connected between nodes C+ and C- and the ESR-value in addition to DC-resistance of L1 and R1 sets a 'reference resistance' later RREF = ESR + R(1) + R1. So, RREF is varying between 10 - 11 ohms in the case of capacitors having maximum ESR of 1 ohm.
The operational amplifier X2 gnerates pulses, which triggers oscillator (X1). If the feedback paths are in balance (=resistor ratios R2/(RREF) (VR1+R5)/R16 are equal) that oscillation is continuous. The pulses of the oscillator are detected with X3 which drives a LED. In the other words, by adjusting a proper feedback (using VR1) the ESR of the measured capacitor is detected by inspection whether the LED is blinking. The ESR value is obtained indirect manner by observing the VR1 value in that condition. A good component for VR1 is a multiturn high precision potentiometer with a proper knob, see foto below of the ESR-meter.
http://img403.imageshack.us/img403/2593/esrmetersch11nv.th.jpg (http://img403.imageshack.us/my.php?image=esrmetersch11nv.jpg)
Figure 34. A simplified schematic diagram of the ESR-meter
Construction
The schematic of the ESR-meter includes the most of the relevant information required, but I add some comment if somebody really want to build a high performance meter. Supply voltage can be almost anything, only the maximum rating of the OPAMP's is critical. Althought a high voltage generates high pulses to the capacitor under measure, so I suggest to use a lowest voltage which is specified by OPAMP. The supply voltage could be taken directly from 9 V battery, but I found the voltage variation be too high to get milliohm accuracy. Therefore, I use a linear regulator LM317 and some good capacitors to get a clean 9 V supply from a unregulated power supply. The inductance of an inductor L1 is not too critical, but the resistance above L1+R1 (should exactly 10 ohm). In the original guide was proposed to get 14 m copper wire having diameter of 0.3 mm and coil it around screw holders of the plactic box. In the model I used (~10x10 cm) that strategy gave a square shape coil inductor with ~ 45 rounds. After I finished the coil I solder it in series with a trimmer and adjusted resistance to 10 ohm. The ABSULUTE values for VR1, R5, R16 and are not critical, but the ratios. In the original design there were used values VR1=100k, R5=1Meg, R16=11k, but I used slighly different values since I got a high precision potentiometer having value of 4.7kohm. Therefore, I suggest to first select the best potentiometer you can find having value ~ 4.7 k - 100k and scale rest values to get proper ratios: R5=10*VR1, R16=(VR1+R5)/100. In the original design there was a switch changing R5 and R16 values to get different measurement ranges. Whatever values/resistances you will use a trimmers in series will be required to get EXACTLY correct ratios. The accuracy of the feedback paths (resistor ratios) sets the accuracy of the device, so if you accept lower performance you can use standard resistors. A suggested candidate for operational amplifier is TL084/TL074: required four OPAMP's in one IC and good performance.
Calibration and Measurements
After soldering all components to PCB some kind of calibration has to be performed. In the case of meter of high (!) ESR values, say 10 ohms, it is quite easy to make cabiration. A zero ohm ESR capacitor (=in respect of the range) can be easily found, making the zero level calibration an easy task. Moreover, the range adjusting can be performed with well defined resistors e.g. by connecting 4 psc 20 ohm resistor in parallel to get a 5 ohm accurate reference. When making mesurements in milliohm range things get *a little bit* complicated:
- There is no 'short', zero ohm ESR reference, since the point is to measure the best capacitors exist !!!
- Milliohm range resistors are hard (impossible ?) to find and cannot be characterized using standard equipment
- ESR of electrolytic capacitors is highly temperature dependent i.e. touching device in the same time changes results
- Effect of the connections/wiring is dramatic: 4 cm COPPER wire diameter of 0.3 mm equals resitance of ~ 10 mohm !!!
It can be easily seen that calibration and measurements have to be performed very carefully to get reliable results. I calibrated the equipment as follows. At first I set all feedback resistors to correct values with trimmers. The LED is blinking while the feedback resistance is at correct range and therefore the value has to be acquired when LED stops blinking. The knob I used shows the position at the accuracy of 1/100 per turn, which gives in theory reading resolution of 1 mohm (10 turns). I connected all capacitors in the same position by soldering to minimize the uncertainty of lead length and contact resistance. Using a high quality capacitor (Panasonic FK, 4700 uF, 10 V, ESR(max) = 18 mohm) I set the 'zero ohm level', take a look foto below about ESR-meter with capacitor under measurements.
I found it more convenient to set zero in purpose to higher level (by trimmer settings), since the device was operation more robust/linear manner when not set set exactly to zero. Moreover, I was not sure whether I will measure later a capacitor having even lower ESR and it would not be possible to get less than zero values with the meter !! This arrangement (arbitrary zero calibration) shifts the acquired values with a fixed offset resistance Roff. I approximated the offset resistance value of 50 mohm as follows. I measured two similar Panasonic FK capacitors in parallel, which halves the ESR but keeps offset resistance unchanged and Roff can be easily calculated. I obtained a value of 50 mohm, and found the result was well reproducible between different measurement sessions. I also checked the linearity of the equipment by connecting a known resistance series with measured capacitors. I constructed some low value resistors by connecting 0.3 ohm resistors in parallel and by cutting a copper wire (diameter of 0.3 mm) to proper lengths. I found that meter was operating in a very linear manner even resistor level down to 10 milliohm, which was a 4 cm legth copper wire (!!) (see foto below for setup).
http://img133.imageshack.us/img133/5999/pict0258d8kq.th.jpg (http://img133.imageshack.us/my.php?image=pict0258d8kq.jpg)
Figure 35. The ESR-meter setup, Panasonic FK capacitor in series with 30 milliohm resistor
As a conclusion of the ESR measurements: it was not a easy task, but after using hours to examining the problem and making tests I was very happy to the results obtained. It was possible to have a proper accuracy using a home made equipment, part cost of some euros. Also these measurements finally showed that the ESR-value of the old CV-EX capacitors is far away specifications - one of the my assumptions when starting to examine BS-feature.
Afterwords:
I hope someone likes this document, at least for me solving this problem has been quite challenging and even sometimes fun ! I hope that you can help to make this document even better by posting answers releted to 'Most wanted' items.
Changelog:
1.0 Created by SikaRippa (13.4.2005)
1.1 Figure links changed (14.4.2005)
1.2 Temperature effects (ch 4.2) and anti-aliasing etc. (ch 3.1) added (19.4.2005)
1.3 Little modifications (mostly typos) (21.4.2005)
1.4 Major update (30.4.2005, "Hyvää Vappua!")
- Info about many cards added
- GW6800 version 1B (explained, measurement results)
- Ultimate modification added
1.5 Chapter 'Noisy audio measurements' added (still under construction..) (11.5.2005)
1.6 Several little modifications and 'Memory voltage modification' added (17.5.2005)
1.7 Major update (13.6.2005)
- Mystery of C136 & C143 capacitors revealed !! (Ch 2.2)
- More accurate specifications for'Ultimate capacitor modification' (Ch 3.3)
- Temperature measurements Ch 5.4
- Component value determination Ch 5.5
1.8 FAQ-list added, minor changes e.g. 'Enhanced capacitor modification' figure fixed
1.9 Some updates (24.6.2005)
- Free capacitors
- Capacitor information
- Noisy audio
- Comments about switching regulator simulation
2.0 Many updates (1.9.2005)
- Fotographs of different cards
- Capacitors lists updated
- Ultimate capacitor modification/gainward modification rewriten
2.1 Many updates (3.12.2005)
- Gainward identification more precise (figure added)
- MSI cards (foto, comments)
- Sparkle (foto, comments)
- Aopen (fotos, comments)
- "Power supply"-chapter added
- "Ultimate capacitor modification"-chapter updated (figure added)
- "Modification of Switching regulator Feedback"-chapter rewritten
- "RoHs-issue"-chapter added
- "Temperature measurements" (measurement results, comments)
2.2 Many Updates (13.4.2006, yes exactly one year....)
- New information about some models
- AC-current measurements
- AC-voltage measurements
- ESR-measurements fully explained
Super-Credits:
Zadah
Credits (sorry for those I missed, I will update list later ;) ):
Bilabong, Faults, zOU, Michael Shouler,exyxxx, MAXIMUS, Terppi, BrxA, WileNius, Maxxer, AION, maaku, JTG, Mr Q, Tiny, pkuitune, Reptile, faior, Mace, mikkomaa, ZeroHero, stoneheart,dandaman, disciple, fitnessbonk, markiemrboo, sunn, Flip, oysteini, AFA-productions (for cameras and lens ;) ), DeNs, Sidolin, julianus, money2, Godefroy, hilton, Sakapo, Tormentor, Jaxradar, Möösi, zxcv,***Deimos***, xotoy, jabjab, inspire, unrealuniverse, robb-x, jromon, soodent, piaskoon, roby58, Doctor6000, uncy_chris, Firechicken, bertos, hannibal216, OnBoard, Sh0rt, xan, Graversen, Kizmo, Augu, hilton, WarPeR, Mr.Z, anzi80, Reflex86, TuskaMies, Icharus, lettubatman, DJ22, and many others..
References:
[1] http://forum.gainward.de/viewtopic.php?t=721
[2] http://www.3dgamers.com/games/skialpin2005/downloads/
[3] http://murobbs.muropaketti.com/thread.jsp?forum=21&thread=347556&msRange=15
[4] http://www.xtremesystems.org/forums/showthread.php?t=48876
[5] http://www.intersil.com/data/fn/fn9134.pdf
[6] http://www.anpec.com.tw/products/download/APL/APL5331/APL5331.htm
[7] http://www.hynix.com/datasheet/pdf/dram/HY5DU283222AF(Rev.0.7).pdf
[8] http://www.low-esr.com/electcompare.html-ssi
[9] http://www.chemi-con.com/u7002/table2.pdf
[10] http://www.nichicon-us.com/english/lib/alminium.pdf
[11] http://forum.msi.com.tw/index.php?topic=78200.new#new
[12] http://www.semicom.co.uk/manufacturer/sanyo/sanyo_oscon/oscon_products_svp.html
[13] http://www.secc.co.jp/pdf/al_e/E15.pdf
[14] http://www.chemi-con.co.jp/Welcome_e.html
[15] http://industrial.panasonic.com/www-data/pdf/ABA0000/ABA0000TE5.pdf
[16] http://www.chemi-con.co.jp/pdf/catalog/al-e1001e/al-pxa-e-050527.pdf
[17] http://www.intersil.com/cda/deviceinfo/0%2C1477%2CISL6534%2C0.html
[18] http://www.newark.com/services/rohs/documents/RohsTechManual_v2.pdf
[19] Ray Porter, "Simple ESR Meter for Electrolytics", TELEVISION Servicing Magazine January and April 1993
------------------------------------------------------------------------------
Most Wanted list:
1. Information about BS-cards not mentioned above (brand/model/codes/etc. or foto)
2. Any comments about this document, especially about presented BS-theory.
3. Some comments from those really involved with this problem, like R&D-team of Gainward. Behave like an real engineer, and give us correct explanation about the problem (if you know !?!?)