all vmod for all 6800

(for all: 3D, throttle, & 2D)
1.1V: GPU = 350Mhz.
Higher either froze or produced artifacts and throttled back down to 300 (2D). Memory artifacts at 950, and even 900.
1.2V: GPU = 370Mhz.
Higher throttled and produced artifacts.
1.3V: GPU = 435Mhz (highest I've finished GT1 before), but normally 425.
1.4V: GPU = 470Mhz???
950 on memory gave artifacts. Even 900. After about 435-440, still 100% artifact free video, but much lower frame-rates. BUT, rivatuner still showed the same high clockspeed throughout. Increasing Mhz higher, and still no artifacts or freezing, running perfectly fine, and fps increases but significantly lower than normal (ie 3DMark01 Nature: 120fps vs 180fps). Now, here is the kicker. I lower the Mhz back down to 400Mhz: fps=100. I lower Mhz even lower to 350Mhz: fps=80. ITS STUCK. Even re-loading the driver doesnt fix it. What could possibly be causing this??

Throughout all 1.4V testing temp was <63C. At 470Mhz, fps dipped again, and any higher Mhz would produce artifacts or freezing. Its like the GPU is disabling pixel shader quads, one after another, me thinks.

Quote:

---

Originally Posted by ***Deimos***

To add video memory usage to your rivatuner graphs:
1. Open Rivatuner. On main tab, in Target Adapter field click next to the 'customize...'
2. Select Hardware Monitoring (2nd from the right)
3. Click Setup, and then in the 'Hardware Monitoring Setup' window, click Plugins in bottom left.
4. There is a list of .dll, you should have some already selected like NBThermalDiode for temp reading... click next to VidMem.dll to add it and select OK.
5. Finally, you have to add it to the graph. Back in the 'Hardware Monitoring Setup' window, add them in the 'DataSources' list by checking them.. click OK.. you're finished!

---

Thanks for good instructions ! I'll test that later myself.

--

Quote:

---

At first when I did that testcase, I thought all my problems were finished. I almost got through GT1 at 440!! (still at 1.316V) However, next morning same problems as before. It truely is bizarre.. one hour everything is running super fine at high overclocks and everything, then boom.. and 410, 400 even 390 throttle.

Incidentally, I also changed the GPU throttle slowdown and Ambient shutdown temps both to 85C.

EDIT: I'm about to go back to 1.2V BIOS, and see what happens then.. haven't tried that in long time, and I'm getting buried alive in dozens of excel spreadsheets of (often contradicting) data.

---

Too bad, you could not make your card (fully) working. I think you have been testing/tweaking so much that nobody is able to give solid answers how to fix problems.

Quote:

---

Throughout all 1.4V testing temp was <63C. At 470Mhz, fps dipped again, and any higher Mhz would produce artifacts or freezing. Its like the GPU is disabling pixel shader quads, one after another, me thinks.

---

GPU disabling pixel units (as a funtion of temperature ??) does not sound a bad theory at all, since your got so huge instantenous FPS drop. In order to proof that some info about internal structure of the GPU would be nice... but I wonder nVidia does not provide detailed schematics of internal blocks.

In an ugly twist of events, I currently have the worst result yet.

It all started this morning. If over-heating is a problem, how to fix... ahhaa... OCZ P4 Eliminator Passive 1U 504g copper heatsink! SO, I put the computer case upside down, put a fun up on top of the heatsink, and away I go. Initially, I had higher temperatures, but it did not throttle.. not once (at 420/900, about 10 runs). Only when I got to 430 would it throttle. Interestingly, I wasnt able to run 950 (ex 370/950).. hmm.. strange. But, I found out that after I clock to 370/750 and then to 370/950, it works (like a charm). However, after 2-3 Nature loops, it would "throttle". Now, although rivatuner shows the same clockspeed, the performance was lower. Not surprising. But, interestingly, after reseting again (by downclocking to 750), it would throttle, even with 30, and even 60s cooldown breaks in between. In fact, It failed even sooner. Its like its on a timer!

trying to get better heatsink contact, I once again found out how difficult it is to get good contact. I continusuly monitored the idle temp... typically ~40-42C. A little tip to left, and its up to 50, a tip to the right and its zooming up to 70, a tip down etc... Here is another "magical" discover. I grasped the heatsink with both hands and squeezed the card. Temps plunged down to 30-32 level! I was able to replicate this a few times. But, never worked with one hand... and I also swear I felt a bit tingly in the fingers (which were moist with sweat). Maybe I accidentally altered the temp monitor signal with sweat on fingers? Or could it be that I had better heatsink contact? I quickly dispelled the 2nd theory.. using foot to operate the mouse, I quickly had artifacts at just 420. Once or twice the screen just went black during this repositioning.. I presume this was the safetely thermal shutdown on the GPU.

So, fun time is over, and I decide to replace the NV5. While cleaning off the AS5 on the BGA's, I get some ont he PCB.. pretty soon there is AS5 everywhere.. so I get the olde 'Precision Tools Contact Cleaner' again.. and spray under the BGA's and under the GPU and use paper towels and cotton swabs to clean up. I do a thourough job, and then dry for a few hours. Now, putting NV5 back on, and try Nature.. strange artifacts (like never before). I clock down.. still there. Internal driver test fails, even at low clocks. rthdribl is freezing every few sec. I flash BIOS to 350/700. Internal driver test still fails. Oh oh! I get same strange artifacts (not like ever seen before). Measure Vgpu = 1.312V, Vmem= 2.71V. Whats the problem? Long story short, I finally get it kind of working after disabling vertex shader (hardware masked one). 350/700 and driver test passes. Coolbits autodetect 375/750. Even test passes up to around 430/800. But, after playing BF2 for a few hours (370/900) when I tried to alt-tab back to desktop, it was all garbly gook.

I think I may have damaged the card during cleaning or when the monitor turned off during repositioning of the heatsink.

seeing all this problems maybe the unlocked pipe was never a good one.
also "contact cleaners" aren't that agressive to destroy cards. I've used it on cpu's, vga's and motherboards and nothing died (well, there did die some, but not from using contact cleaner).
Maybe u cracked the core a bit. I have had a stupid kill in the past when cleaning a 1700+ with kitchenpaper. The paper was to 'hard' and so I cracked some very small pieces from my Atlons core.

***Deimos*** quite complex story .. hard to figure out all testings/modifications you have done now and before, what might happened during each step and why.

I suppose your extra pipes/vertex units were fully working until now ? Even at highest OC-frequencies ? If you are not able to get those working anymore (even at stock frequencies) it is clear something has happened. The contact cleaner (+ thermal paste between BGA balls due to cleaning ??) could be the reason or squeezing of the heatsing & card. However, it does not sound likely that squeezing would broke ONLY one fully working vertex unit. Therefore as wittekakker stated below, maybe the locked vertex shader was originally somehow out of specifications and now broke down for some reason (temperature etc.)

16/5 and still throttling after 2-3 runs. and giving 1-2 minute in between for cooldown just makes it worse (you have to see it to believe it)

so , this morning, I'm fed up with the lower fps from 16/5, and use rivatuner to go all the way to 16/6. Reboot.. everything seems fine. Open rivatuner hardware monitor and do some benchmarking. Benchmark scores are up, artifacts are gone, and you're looking at one happy camper. I was testing 420/950. I already did a whole 5th benchmark run at this speed.. wow.. magically throttling has fixed itself. But surely 6th run will fail right? Nope.. goes all the way through without a hitch... I run like a crazy monkey to my brother to tell him the good news.. he starts trying to punch me because I'm disrupting his BF2 gameplay... so about 5 min later, I come back to the computer, I click benchmark just for the sake of it... instead of 160-162fps, its now 128-130fps.. BUMMER. its back.

This morning, after boot up, first 2 are fine, but the third time it freezes.. well at least those ugly artifacts made themselves disappear. Rivatuner shows clocks throttling. Back to the same old.

EDIT: The three constants in ALL my tests so far. Athlon64, ]b]NF3[/b] and this 6800. Still cant convince my brother to slide my 6800 in there... he's absolutely certain its possessed, and that the evil spirits will escape into his precious system.

EDIT2:
http://www.xfxforce.com/forums/threa...D=18518&#18518

Just re-read the thread (its kinda long, no?):

I read up on the voltage regulator IC.. its really neat. Unfortunately, increasing the resistance to F_SYNC would mean cutting traces, since to gnd = default 300KHz, and 10K = 1000KHz for switching frequency.

SikaRippa, your thoughts?
TP1 to gnd - stabilize GDDR: good idea?
TP11 to gnd - stabilize Vgpu: good idea

Quote:

---

(Card 2:
before, 400/700Mhz and black screen in most of the newest games (like HL2)
added 1x3000 uF cap to TP1 and 1*3000 uF soldered paraller to the "core" caps, they are those 2,5V /1500 uF caps on lower part of the card.
now running 430/800 Mhz without any Black screens and with 16x1,6 pipes..
getting 11250 points in 3dmark03

Ps. You should not solder any caps between TP11 and GND because TP11 is not directly connencted to Core rail..... YOu should only solder those extra caps directly to the pins of core capasitors....
)

---

parallel to which one?

I also have a special PSU that is normally only used for hard drives. It has 400W PSU, and I am thinking of using it seperately to power video card.. any thoughs or comments? Synchronization problem maybe, or nothing to worry about??

...

Quote:

---

Originally Posted by ***Deimos***

...evil spirits will escape into his precious system.

---

.. not a very scientifical explanation, but maybe the best so far: the number of WIERD problems with your card is so huge.

Quote:

---

EDIT2:
http://www.xfxforce.com/forums/threa...D=18518&#18518

---

Broken link, forum does not exist anymore (?)

Quote:

---

Just re-read the thread (its kinda long, no?):

---

Yep, long enough, I have not myself read all post (>90 %)

Quote:

---

I read up on the voltage regulator IC.. its really neat. Unfortunately, increasing the resistance to F_SYNC would mean cutting traces, since to gnd = default 300KHz, and 10K = 1000KHz for switching frequency.

---

Idea to increase the oscillation frequency is DECREASE the resistance between FS_SYNC and GND, take a look the datasheet: lower resistance = lower PULSE WIDTH (=higher frequency) !! I investigated the problem little bit. I measured resistance between FS_SYNC and GND and got a value ~ 0 ohm. That is a quite odd result:

- The frequency IS 300 kHz (I have measured)
- Having standard frequency (300 kHz) means FS_SYNC should be left open i.e. high resistance
- A low resistance should indicate a VERY high frequency, say 1 MHz

Could it be possible there is some kind of pull-down inside the chip or maybe the small SMD resistor connected to between FS_SYNC and GND has a low value ?? Anyway, since the output frequency IS 300 kHz in default the frequency should be altered with suitable added (variable) resistor. There is mentioned in datasheet that any extra capacitance to that node should be avoided, so mod should be done carefully (short leads etc.). In any case I do not suggest to do this mod WITHOUT having an oscilloscope to test the frequency all the time.

Quote:

---

SikaRippa, your thoughts?
TP1 to gnd - stabilize GDDR: good idea?
TP11 to gnd - stabilize Vgpu: good idea
parallel to which one?

---

A GOOD (=low-ESR) added capacitor to both those nodes should give some boost to performance. Long wires (even some centimeters!!) and bad soldering ruin the mod easily.

GDDR: TP1 is not good testpoint, use TP10 & TP9 because this is electrically and mechanically much better alternative.
http://img69.imageshack.us/img69/4061/capmod35gz.th.jpg

Vcore: Yes, you should solder capacitor directly to the leads of purple "1500 uf/2.5 V" SVP capacitors instead of TP11, see below:
http://img250.imageshack.us/img250/1...lev14qv.th.jpg

Quote:

---

I also have a special PSU that is normally only used for hard drives. It has 400W PSU, and I am thinking of using it seperately to power video card.. any thoughs or comments? Synchronization problem maybe, or nothing to worry about??

---

There are severe risks to have two different PSU's to drive the same system. The problem is DC-voltage level, which can differ between two PSU's. If (when) DC-levels differ current flows between PSU's !! In theory if you connect PSU's in the same grouded AC-outlet there should not be any problems BUT measure DC-voltage between GND's before !!!

What would an ideal voltage for my memory be? I want to use a fixed resistor, with a safe slightly higher vmem.

What ohm resistor should I use?

2.9v be ok?

SikaRippa.. are you sure about what you said about FS_SYNC. I have the pdf open here in front of me and it says: ground FS_SYNC = 300KHz (default), use 10K resistor for up to 1000KHz, if left open (infinite resistance) it will default to 300Khz.

Thus, the 0 ohm you measured and 300Khz is perfectly normal.. how its supposed to be. I would need to increase the resistance to increase the switching frequency.

EDIT: I've rulled out the 2nd PSU idea. A bit dangerous, but above all, too impractical. There is already hardly any room for just 1 power supply in my case.

EDIT2: I am working on a heatsink to cover the 5 transistors/IC at the bottom of the card. I think I might have forgot to mention but on my card, one of those 8 leg IC (presuably power MOFSET) at the bottom in instead packaged in 4 pin on one side low-profile design. I'm thinking perhaps replacing these things might do wonders.. certainly better than doing something crazy like 2nd molex connector replacement mod.

Quote:

---

Originally Posted by bigboi86

What would an ideal voltage for my memory be? I want to use a fixed resistor, with a safe slightly higher vmem.

What ohm resistor should I use?

2.9v be ok?

---

The 'optimum' seems to dependent on what kind of memory chips you (=how lucky you are). Some cards OC's quite high even with stock voltages while some do not give any performance increase even at very high voltage level.Anyway, I think the values below 2.95 V are 'safe' or not too risky.

I put the link to mod (again):

http://img216.imageshack.us/img216/4...mod16ir.th.jpg

The required resistor value can be acquired from voltage/resistor graph, e.g. resistor value 2200 ohm should give ~ 2.9 V

Note !! The actual resistors in card (due to tolerances) changes so MEASURE obtained Vmem value.

Quote:

---

Originally Posted by ***Deimos***

SikaRippa.. are you sure about what you said about FS_SYNC. I have the pdf open here in front of me and it says: ground FS_SYNC = 300KHz (default), use 10K resistor for up to 1000KHz, if left open (infinite resistance) it will default to 300Khz.

Thus, the 0 ohm you measured and 300Khz is perfectly normal.. how its supposed to be. I would need to increase the resistance to increase the switching frequency.

EDIT: I've rulled out the 2nd PSU idea. A bit dangerous, but above all, too impractical. There is already hardly any room for just 1 power supply in my case.

EDIT2: I am working on a heatsink to cover the 5 transistors/IC at the bottom of the card. I think I might have forgot to mention but on my card, one of those 8 leg IC (presuably power MOFSET) at the bottom in instead packaged in 4 pin on one side low-profile design. I'm thinking perhaps replacing these things might do wonders.. certainly better than doing something crazy like 2nd molex connector replacement mod.

---

I was *little bit* confused for a while, but now I got it:

There are two versions of the datasheet (I was reading a print copy of the old one) and there are different information in those versions !!! I cheched out the new version and there is indeed mentioned that a short also sets output to 300 kHz. In the first version only an open condition was mentioned..
I wonder the short is done with the SMD resistor (zero ohm) besides the FS_SYNC pin, so if you want to modify frequency remove that and put e.g. 10 kohm instead.

Replacing MOSFET's to better ones is good idea but be sure they are electrically equivalent (except lower Ron). For instance a higher gate capacitance would increase the heat generation inside regulator chip due to higher load.

Quote:

---

Originally Posted by SikaRippa

The 'optimum' seems to dependent on what kind of memory chips you (=how lucky you are). Some cards OC's quite high even with stock voltages while some do not give any performance increase even at very high voltage level.Anyway, I think the values below 2.95 V are 'safe' or not too risky.

I put the link to mod (again):

http://img216.imageshack.us/img216/4...mod16ir.th.jpg

The required resistor value can be acquired from voltage/resistor graph, e.g. resistor value 2200 ohm should give ~ 2.9 V

Note !! The actual resistors in card (due to tolerances) changes so MEASURE obtained Vmem value.

---

So I try putting a 2.2k ohm resistor between TP13 and TP14 and measure voltage, correct?

Quote:

---

Originally Posted by bigboi86

So I try putting a 2.2k ohm resistor between TP13 and TP14 and measure voltage, correct?

---

If you only have fixed value resistors, sure.
However, it certainly is more convenient to use a variable resistor (pot). Ideally, putting something like 10K pot in series with 2K fixed resistor, as mentioned in SikkaRipa's guide.

Quote:

---

Originally Posted by ***Deimos***

If you only have fixed value resistors, sure.
However, it certainly is more convenient to use a variable resistor (pot). Ideally, putting something like 10K pot in series with 2K fixed resistor, as mentioned in SikkaRipa's guide.

---

I went ahead and popped a 2.2k ohm resistor between TP13 and TP14, I don't have a DMM right now, but I believe it's somewhere around 2.8-2.85v, because the ram will run stock timings up to about 900mhz.

I think it needs a bit more voltage, I might have to do the variable resistor trick, even though my soldering skills aren't very good.

I got a 10k trimmer and a 100k trimmer sitting here.. I don't know which side makes the resistance go up.. or how to wire it up ...

I'm a noob when it comes to voltmodding :p:

Quote:

---

Originally Posted by bigboi86

I went ahead and popped a 2.2k ohm resistor between TP13 and TP14, I don't have a DMM right now, but I believe it's somewhere around 2.8-2.85v, because the ram will run stock timings up to about 900mhz.

I think it needs a bit more voltage, I might have to do the variable resistor trick, even though my soldering skills aren't very good.

I got a 10k trimmer and a 100k trimmer sitting here.. I don't know which side makes the resistance go up.. or how to wire it up ...

I'm a noob when it comes to voltmodding :p:

---

I suppose you have around 2.9 V right now, if the tolerances of components are at normal range. I think 2.9 V is a *safe* maximum, if you consider to go higher test the temperatures of the MOSFET's e.g. carefully with finger (little 8 lead chips back of card, see foto below):

http://img363.imageshack.us/img363/4...782112x.th.jpg

.. if those are very hot some extra cooling would be nice.

--

You have 900 MHz now .. how about maximum before modding ?

Quote:

---

Originally Posted by ***Deimos***

Just re-read the thread (its kinda long, no?):

I read up on the voltage regulator IC.. its really neat. Unfortunately, increasing the resistance to F_SYNC would mean cutting traces, since to gnd = default 300KHz, and 10K = 1000KHz for switching frequency.

SikaRippa, your thoughts?
TP1 to gnd - stabilize GDDR: good idea?
TP11 to gnd - stabilize Vgpu: good idea

parallel to which one?

I also have a special PSU that is normally only used for hard drives. It has 400W PSU, and I am thinking of using it seperately to power video card.. any thoughs or comments? Synchronization problem maybe, or nothing to worry about??

---

I went around the city, and found 2 electronic parts stores. However, by the time I got there, they were closed. :(

However, I tried a 2200uF capacitor:
TP11-TP4: seemed to make things worse. Instead of just throttling, I also had blank screen. Didn't help stabilize GPU at all.
TP11-TP9: tried TP9 as different ground point. No go. Same worse behavior.
TP1-TP4: As alternative to TP10-TP9, some users suggested using TP1 to also stabilize the memory. It might be working. Doesnt seem to be making it worse. I couldn't get 960 stable (anymore), but perhaps thats due to higher temperature in my room.

-----
I found a really nice chipset heatsink that will fit over the 4-5 voltage regulators and FETS at the bottom corner of the card. However, the big problem is that they are all different heights, and the TSOP chips are lower profile than the surrounding capacitors.. I have some ~30tau Al shim, and ~10tau steel shim. Unfortunately, its about 20tau difference between the height of TSOP and those FETS.

Suggestions? Just put lots of silicon grease? I really dont want to use double shims. Better to use thicker shims and better contact with FETS or thinner shims and better contacts with the TSOPs?

Quote:

---

Originally Posted by SikaRippa

I suppose you have around 2.9 V right now, if the tolerances of components are at normal range. I think 2.9 V is a *safe* maximum, if you consider to go higher test the temperatures of the MOSFET's e.g. carefully with finger (little 8 lead chips back of card, see foto below):

http://img363.imageshack.us/img363/4...782112x.th.jpg

.. if those are very hot some extra cooling would be nice.

--

You have 900 MHz now .. how about maximum before modding ?

---

The mosfets were hot before I even modded it lol.. they are so close to my RAM slots I will have to have an UBER low profile ramsink or something.

The max I could get before the mod was about 850. Not that great of an increase, but pretty decent IMO.

Quote:

---

Originally Posted by bigboi86

The mosfets were hot before I even modded it lol.. they are so close to my RAM slots I will have to have an UBER low profile ramsink or something.

The max I could get before the mod was about 850. Not that great of an increase, but pretty decent IMO.

---

do you mean the ICs nearest to the power connector or the ones at the bottom of the card where the AGP tab is (which would be near your DIMM slips)?

The first, I think are used for Vmem (for the memory), and the ones at the bottom are used for Vgpu.

Personally, on my board, the ones at the top (as shown on SikaRipa's pic) hardly get warm, even at 950+. They did get a bit hot after voltage mod to 2.9V, but it passed the "hold finger there for 10s, so its about 60C test".

the ones on the bottom thought.. get very hot. It might be my imagination by one of the FETs (left) seems to be more scorching hot than the other (both are >>60C without fan)

Quote:

---

Originally Posted by bigboi86

The mosfets were hot before I even modded it lol.. they are so close to my RAM slots I will have to have an UBER low profile ramsink or something.

The max I could get before the mod was about 850. Not that great of an increase, but pretty decent IMO.

---

Yep, cooling of ALL those MOSFET's (+ regulator) is not bad idea at all. I measured tempetures of those chips (and posted results some pages ago in this thread) and they all run hot even stock voltages.

--

850 MHz - 900 MHz : not a world record, but at least the level specified to those memories. All chips are unique and it does not help to give more voltage while having 'slow' or 'typical' memory chip.

Quote:

---

Originally Posted by ***Deimos***

I went around the city, and found 2 electronic parts stores. However, by the time I got there, they were closed. :(

However, I tried a 2200uF capacitor:
TP11-TP4: seemed to make things worse. Instead of just throttling, I also had blank screen. Didn't help stabilize GPU at all.
TP11-TP9: tried TP9 as different ground point. No go. Same worse behavior.
TP1-TP4: As alternative to TP10-TP9, some users suggested using TP1 to also stabilize the memory. It might be working. Doesnt seem to be making it worse. I couldn't get 960 stable (anymore), but perhaps thats due to higher temperature in my room.

-----
I found a really nice chipset heatsink that will fit over the 4-5 voltage regulators and FETS at the bottom corner of the card. However, the big problem is that they are all different heights, and the TSOP chips are lower profile than the surrounding capacitors.. I have some ~30tau Al shim, and ~10tau steel shim. Unfortunately, its about 20tau difference between the height of TSOP and those FETS.

Suggestions? Just put lots of silicon grease? I really dont want to use double shims. Better to use thicker shims and better contact with FETS or thinner shims and better contacts with the TSOPs?

---

TP11-TP4 (or any ground) is NOT optimal/good way to add capacitor for Vcore as I explained above. The resistance between TP11 and Vcore capacitor +node is ~1 ohm or at least far too high. ESR of the original SVP-capacitors is extremely low ~10 milliohms each. If you put a capacitor (not matter how good) you have at least ~ 1 ohm extra ESR due to wiring inside the card. Is the 1 ohm much ? In normal case not, but here it totally deteriorates the effect of the added capacitor. The purpose of those capacitors is to provide several amperes of current within nanoseconds: you can imagine that such current cannot be driven thorogh 1 ohm resistors...

TP10 is much more better than TP1 for Vmem capacitor. Basically the same reason as above. TP1 is mechanically long away from proper place and extra series resistance will be added. That has been proofed: some BS-affected cards that were not able to repair with TP1-mod could make it with TP10-mod.

--

Heat problem. I suggest to use shims which give best contact with MOSFEt's since those are the hottest.

what do you suggest for vmem?

running now at 2.95v and about 960mhz (still testing). fans on front and back of the card so nothing is going to overheat.

Quote:

---

Originally Posted by ya_master

what do you suggest for vmem?

running now at 2.95v and about 960mhz (still testing). fans on front and back of the card so nothing is going to overheat.

---

Sweeet mem clock.

I need to crank mine up a lil more. 950mhz+ would be nice.

i would like to crack the 1ghz or even more but it looks like i'm gonna need more vmem for that. and i heard it's not recommended to go above 2.95v.

it is not doing any good to run 1ghz at 3.xxv when the card gets crispy toasted after a week right? ;)

a package of small heatsinks is going to arrive tomorrow. i will put them on anything feeling warmer than my finger. that should do the trick.

and it looks like i will have to redo the caps mod. gotta read through the whole thread. damn. :(