!! Version 2.0 of document posted !!
- Finally !
- 'Ultimate capacitor modification' succesfully tested and guide rewritten
- Many fotos of cards added e.g. good quality foto of GW6800 'dual fan'
--
I have plenty of more material, but I want to put this update NOW since many users have requested a better version for 'ultimate capacitor modification'. In next update I'll have e.g. more measurement results (oscilloscope fotos etc.)
Quote:
Originally Posted by SikaRippa
Hello, i thought i'd update you on my LE, working nearly 4 months BS free :woot:Quote:
Originally Posted by sunn
This time i included some proper size pictures
<a href="http://img89.imageshack.us/my.php?image=aopen6800le17gc.jpg" target="_blank"><img src="http://img89.imageshack.us/img89/7651/aopen6800le17gc.th.jpg" border="0" alt="Free Image Hosting at www.ImageShack.us" /></a>
<a href="http://img89.imageshack.us/my.php?image=aopen6800le22nv.jpg" target="_blank"><img src="http://img89.imageshack.us/img89/3430/aopen6800le22nv.th.jpg" border="0" alt="Free Image Hosting at www.ImageShack.us" /></a>
some news.
good: at 700 MHz no BS anymore
bad: at 800 - BS again... after 30 sec of GTA San Andreas...
i'm very disappointed now.
maybe 2x470uF is too low?
soon i'll recieve 2 caps of Sanyo SVP series (1200uF 4V).
maybe they will be my last chance.
:(
Nice fotos :thumbsup: I am planning to make a chapter '6800(le) cards not (yet?) affected to BS and these fotos would be good for that purpose.Quote:
Originally Posted by sunn
--
I found something special in those figures. Take a look this figure:
http://img388.imageshack.us/img388/3...ison1bp.th.jpg
.. where is zoomed area from Aopen and GW-cards. See the difference (size) of the ceramic filtering capacitors ?? The size corresponds to capacitance value in the case of the ceramics (in this case). This can be one good reason your card (model) is not giving BS. Moreover, the same kind of (larger) capacitors seem to be inserted many places at the bottom part of Aopen (all connected to Vmem!).
Aopen had also version, like this:
http://aopen.jp/products/vga/large/A...00-DV128lg.jpg
having 'normal' sized ceramics (bottom of green caps)
It is pity you can not get more 'power' :( . Maybe, your card has 2.8 ns memories (have you checked codes?). Larger caps might help, at least many users have get some (not too much in every case) increase to clocks with cap mod.Quote:
Originally Posted by zxcv
Quote:
Originally Posted by SikaRippa
yes. card has hynix 2.8 ns memories. but i can set memory clock to 933mhz.. there are no artifacts at all in some 3d applications..
but serious games and tests (like doom3, 3dmark05, GTA SA) gives BS even at 800 MHZ. without any artifacts.
I'm sure its nice to have 933Mhz working for 2d desktop but if you cant run any benchmark or games, then what's the point. How did you even get to 933? Disable internal clock test in rivatuner?
The cap mod doesn't really help overclocking of memory. I tried 2700uF and 4700uF, on both locations (TP1 and TP9), and at best only 5-10Mhz more before onset of artifacts.
However, I do believe that it helps people who are having problems with black screens which might be caused by insufficient power to the volt reg for memory. From the accounts SilkRippa mentions, it seems to be helping some people. Fortunately for me I have the 570uF caps, and dont have any black screen at all... if you do, please try to RMA card instead of toying around with it and possibly destroying it.
OK. Your memories ARE capable to high speed (=faster that 'regular' 2.8 ns). The problem is that the switching power supply design of this card is bad and usually (~allways) OC'ing is limited by BS. Some programs run very well, but when all memories are active in the same time card is not capable to provide enough current and shut down (BS).Quote:
Originally Posted by zxcv
In more detail, my theory says that the problem is (see document for deep details) failing overcurrent protection in such case. Added capacitor or better replacement capacitor provided more current sourcing capability and card survive 'short' condition for a longer time.
In some cards (like new GW's, version 1b) they have modified the feedback loop of the switching regulator (see document) to have a slower response to shorts. Therefore these new models are working better even having original capacitors. I am planning to examine this modification in detail, since I have now measured feedback component values for BS and non-BS cards. I wonder that this modification will help some users to have even more OC'ing capability or remove BS. This mod might (when invented in the first place ;) )be more hard to do, since soldering/desoldering of tiny SMD components will be required.
zxcv: Are you interesting about that kind of Extreme modding ? Do you have a multimeter with capacitance measurement function ?
I considered this little bit also previous post..Quote:
Originally Posted by ***Deimos***
--
My aim has been BS removing, but many users that have a BS-affected have reported all cases some increase in OC-capability (up to 50 MHz I think), because operation has been BS limited. Cards that are NOT BS-affected do not gain so much with cap modding. In your case (XFX version 2, Sanyo SVP 560uF ? capacitors) it is evident that improvent is only minimal: your have more capacitance for Vmem at the beginning and those caps are superior quality when compared to 'green capacitors' or CV-EX's.
SilkaRippa it sounds you know what you are doing.. but I think adding relatively large capacitor (like coupling/bypass cap) from vmem to ground is insufficient bandaid solution. Doesn't this adversely affect low-pass performance (cap to ground)? I'm also worried about the toll on the vreg of charging such large cap. Ultimately, I think you are right about improving the feedback loop. I'm not a practicing electronics so a little rusty, but if the problem is indeed spikes in current drawn, wouldn't an inductor be more appropriate to reduce the inductance in the circuit?
Yes, adding a large capacitor to Vmem node is a brute force method and very large value can cause a number of problems: card might not start up or feedback loop transfer characteristics changes too much (=>many problems).Quote:
Originally Posted by ***Deimos***
However, this mod (against BS) really works for the most users even when having cap as large as 4700 uF. I wonder, that usually quite large added capacitors do not cause problems because the long leads (=several mm at minimum) and moderate ESR value cause a higher impedance than due to original capacitors.
If the switching regulator circuit would re-designed I would suggest several changes. For instance the inductance value is far too low (1 uF), since lower inductance means higher current peaks (but faster response => trade-off). Also the input voltage -output voltage ratio is very low and output power level high - and varying extremely rapidly. (I have new measurement results (not yet fully analysed), that shows how large is the current level and how rapidly it is changing.)
In theory, using aqcuired data, an optimal set of components could be calculated. In practice following aspects make it a HARD task:
- Parasitics (extra resistance/capacitance/inductance) of components and PCB, has considerably effect to response
- Load due to memories is hard to model i.e. current min/max and risetime/falltime
- For a standard user it is too hard task to obtain a number different special components and solder those e.g. small SMD capacitors are extremely difficult to change
Therefore, I have mainly considered modifications that are simply to understand and easy to perform. Anyway, when I get some time I'll investigate the feedback modification more carefully - and report the results.
Yes, i'm very intrested.Quote:
Originally Posted by SikaRippa
Unfortunatly my home multimiter cannot measure capacity. But I'll try to find such multimeter.
Also i have good soldering equipment on my work. so... soldering of smd-components (capacitors) shouldn't cause any troubles =)
Fine ! I'll have to think/simulate/analyze about that very carefully - take some time, cannot promise anything yet.Quote:
Originally Posted by zxcv
raising R or lowering C values in FB2 RC-circuit should help avoid wrong shorts...
http://img121.exs.cx/my.php?loc=img1...ge=fig78fq.jpg
Yes.. but how about response to whole transfer function and what are the correct/optimal values.. . The desoldering/soldering SMD's is so nasty to perform that nobody want to do it several times. Moreover, tiny solder pads tend to 'burn' useless after few/some attempts.Quote:
Originally Posted by zxcv
--
I had nice ~ 2-3 hours while I analyzed component values (new ones got from a badly 'BS' card) and compared to 'GW version 1b. I use the same kind of method described in Ch 5.6. The components under interest can be located using a figure below:
http://img148.echo.cx/img148/2742/memfb17hp.th.jpg
I confirmed that *them* have CHANGED values for capacitances but AFAIK the resistor values are the same. Using an iterative trial-and-error method I found (!) values for those components with acceptable accuracy. I wonder (hope) they have used standard values so the obtained value can be rounded to nearest 'correct' one.
But, please be patient 'zxcv' (and everyone willing to perform this mod). I'll give those values after I have made electrical simulations and saw that those make sense. I want to really see what they have done, before giving detailed instructions. (I am too tired to do simulations tonight...)
I wish I knew one of the pcb designers.. it would be so much easier just asking them.. but probably not as rewarding.
I Sika, i received the card on saturday... its working fine!
i will reinstall my system now and make some benchs... then i will public the table with the oc results...
fyi money2 ( i had to answer public cause pm-systeme isnt working on my machine at work... ) :)
Yep, I would be great to heard ANY comments from those are responsible of designing/modifying this card. I am sure someone of those have read/see this document/thread, but the are not allowed (nor willing) to say anything.Quote:
Originally Posted by ***Deimos***
I was talking this issue with moderator of GW's forum. He confirmed that it is a 'design flaw' and GW has a fixed version version, but NO comments about identifying the faulty card (or what they have modified). You can not find it anymore, since the thread was deleled very soon after those comments, why ? Maybe he was not allowed to say anything about design flaw in public forum..
'zxcv' or somebody else interesting about 'Xtreme'-modding of GF6800 cards..
I have simulated the AC-response of switching regulator (memory part) with an iSim-simulator. I am not going to explain (in this post) detailed what this simulation electrically means, only show the results.
Component labels indicates those shown in this figure:
http://img148.imageshack.us/img148/2...mfb17hp.th.jpg
I acquired following values (see post #1 for details about HOW):
BS-affected card:
C1 = 40 nF (~ +/- 10 nF)
C2 = 33 nF (~ +/- 10 nF)
C3 = 120 nF (~ +/- 20 nF)
Non-BS-affected card:
C1 = 90 nF (~ +/- 10 nF)
C2 = 15 nF (~ +/- 5 nF)
C3 = 135 nF (~ +/- 20 nF)
.. not a big deal, does not affect on anything ???? Wrong, those little changes DO have an effect see the simulation below:
http://img138.imageshack.us/img138/4...sac05gi.th.jpg
So, the response of the non-BS card is faster i.e. the regulator has (more) gain at higher frequencies !! Propably this is enough to remove BS in certain conditions. BUT (a BIG BUT) this mod do not remove the high supply current variation (and ripple-current) and therefore BS can appear also with these cards - but it takes longer time.
--
How to do this mod ? Is it sensible to try it out ? For a highly qualified solderer this mod is piece of cake, but for standard user I do not recommend: soldering of tiny components IS difficult. If properly done it can be very usefull, BUT I CANNOT GUARANTEE ANYTHING SINCE THIS HAS NOT YET TESTED BY ANYONE.
--
Before you (zxcv?) try this out (I know you will anyway ;) ) Please do following test: Measure capacitance between 'n4' - GND and 'n3' - GND (GND to - terminal of meter !!!) and post the values (see figure above for those points) I will analyze values to find out if your card is similar to BS-card I have measured.
I have a GF Sparkle 6800 and I have problems with BS when playing HL2 and CSS. The suggestions I have got so far is to get a new PSU. The one I have is an AOpen 300 W which can deliver 15 A och the 12 V line. Should I get a better PSU or should I modify the card with the capacitor as described in this thread?
Hi,Quote:
Originally Posted by xotoy
Power supply has only a minor effect i.e. having very good supply BS will in some conditions disappear (becomes more rare). Usually the effect is so minimal, that new power supply cannot consider as a solution for the problem.
Easiest and most 'safe' mod is 'Enhanced capacitor mod', which in most cases removes BS. For some cards this mod does not work, and those 'green caps' should be changed (this mod has been given very good results with 'totally BS' card)
--
Is your card similar to this one:
http://img206.imageshack.us/img206/4...00x19rt.th.jpg
.. and it really is 6800 (not 6800le) ?. Could you specify any version code etc. for your card and the labels for 'green caps' (e.g "4 3 470 6E") :)
Thanks for your reply. It is a 6800 (NOT le). I will take closer look at the card and also supply a picture. I did change the cooling unit to an AC, do you want a picture of the original cooling unit as well?
Nice you are ready to provide info about Sparkle, it is rare and very few data available (so far! )Quote:
Originally Posted by xotoy
Fotos are welcome :) also about original cooler if not too hard task..
by the way sika, your thread reached already 15.000 Views :-D
Noticed :D, not bad... interesting issue (?)Quote:
Originally Posted by money2
--
BTW, I sent email for you, got it ?