Nvidia 8800GTS Vmods

[SF3D]

Some useful info.



Example: Set 527MHz core speed= you get 540MHz real core speed and 1242MHz shader clock. Set 554 = 567MHz real and 1296MHz shader clock.

Table is easy to read, when you find out the right way

Now just find your sweet spot

[SF3D]

I'm talking alone here, but that's ok

I just modified my GTS bios and results are promising.

3DMark06 with 1680x1050 4AA 16AF

Shader clock 1188. (default)
06 Score= 6401
Sm2 =2664
Sm3 =2412

Shader clock modified to 1296. (default)
06 Score=6531
Sm2 =2720
Sm3 =2478

[Orion24]

I'm glad to see that overclocking the GPU clock, increases the shader clock by an equal percentage. Can't wait to mod my GTX, but I'm held back by the cooler. I'm yet to find an adaptable waterblock

[railmeat]

I took a risk and flashed it with gtx bios (nvflash -4 -5 -6 to override board id mismatch) it works! At 700/2000 = 203xx on 3dmark 05. but i dont know if the extra stream prosessors unlocked.

wow your a madman,volt modded and flashed with gtx bios.you did amazing.nicely done.i have a evga 8800gts,i sell my cards every 6 months get %70 back and put up another %30 for new 1.i dont really o\c cards,only cpus.dont want 1 of my many freinds who buys my old cards having issues.nice work tho man!!!...nice work ALL OF YOU WOW!!!!

since bf2 came out i had 6800gt,6800ultra,7800gt,7900gtx(on water),7950gx2,now 8800gts,cant wait til 8850-8900 wo0

[ViperJohn]

No I haven't, but it works just like primarion 3535.

I undo my Vmod. Card was too hot with aircooling. I have to mod my swiftech waterblock and watercool this hottie

The pinout has to be different on the PX3540 from the PX3535.

The Vcore mod point traces to pin 19 of the chip and on a PX3535 that is "Out4" which is a mosfet gate drive output.

Viper

[SF3D]

Vmod point is not connected to pin 19. It's connected to pin 26 VSENN.

I don't know why primarion doesn't release datasheet? I would like to see that.

[t024484]

The overclocking table that SF3D has produced, is a bit overloaded with info.

I have re-edited the table, giving an easier and faster information where the sweet spots are.



You can change the Shader frequency in this table by changing the default shader value (with Nibitor/NVflash)
from 1188 to 1188 + X , where X is a multiple of 54.
This X will then be added to all shader values in the table.
X can also be a negative figure, in that case the shader frequency will be decreased.

[SF3D]

Thanks a lot! I was going to do new tables, but you saved my time.

[ViperJohn]

Vmod point is not connected to pin 19. It's connected to pin 26 VSENN.

I don't know why primarion doesn't release datasheet? I would like to see that.

I have two 8800GTX's here and the mod point, which is the same on the GTX and GTS, traces to pin 19. Reads over 15Mohm to 26. Damn odd lol.

Viper

[SF3D]

Well, maybe this explain it to you. Pin 19 and 26 is connected through this circuitry. I can't be 100% sure of that. (Red is vmod point)



Pin 19 gives output voltage, and pin 26 set's that voltage. Resistance between them decides what output voltage will be.

GTX/GTS vmod points are different, but basic idea is same. Pin 19 and 26.

[ViperJohn]

Well, maybe this explain it to you. Pin 19 and 26 is connected through this circuitry. I can't be 100% sure of that. (Red is vmod point)



Pin 19 gives output voltage, and pin 26 set's that voltage. Resistance between them decides what output voltage will be.

GTX/GTS vmod points are different, but basic idea is same. Pin 19 and 26.

The GTX Vcore mods at the same end of the same SMR with a 1k trimmer to grd. (VR-Zone says 500 but a 1K is a better starting point IMO). The components in question are in ever so slightly different locations though but the general layout is the same in the mod area. Your tracing to the left of your red "X" is the same on the GXT although no automated test point pad exists in between the mod target SMR and the SMC to the left (wish it did have that pad to solder to!). The green tracing to pin 26 doesn't exist on the GTX. It dives into a thru-plated hole at the point to an internal layer and I can't find where it pops back up but it sure isn't VsenN (26) or VsenP (27). VsenP was where the 78/7900 cards all modded.

It makes no damn sense I know but the mod works lol. Your wouldn't be able pickup a reference B+ at a mosfet gate drive output as indicated in the 3535 "datasheet" but that datasheet is far from complete. The reference circuit does not show any of the even gate drive "outs" being used at all and normally the odd/even out pairs would be driving the hi/lo (or lo/hi) mosfet pairs. There is a lot more to these regulators than the minimal info we are given in the marketing 3535 datasheet thats for sure.

Viper

[SF3D]

Did some 3DMark runs today. I'm getting black screens after 1.43 Vgpu We really need that datasheet to verify this mod 100%. Or maybe my PSU (Antec truepower 2.0 550W) is not enough.

Scores:
3dMark01: 69094
3DMark03: 35644 (bad run)
3DMark05: 21127
3DMark06: 12422

Good scores for starters

[ViperJohn]

Did some 3DMark runs today. I'm getting black screens after 1.43 Vgpu We really need that datasheet to verify this mod 100%. Or maybe my PSU (Antec truepower 2.0 550W) is not enough.

Scores:
3dMark01: 69094
3DMark03: 35644 (bad run)
3DMark05: 21127
3DMark06: 12422

Good scores for starters

Definately on that datasheet!

What was the die temp at Vcore PS shutdown???

Viper

[SF3D]

Gpu temperature was around 30 celsius.

[ViperJohn]

Gpu temperature was around 30 celsius.

Dang I should have asked this before but did the shutdown occur GPU loaded or idle?

Viper

[SF3D]

Sorry, I could be more accurate.

It appears in most heaviest parts of 3DM06. (GT1 when door lifts in the middle of test, or in GT2)

Vgpu is 0V after that, so it must be that primarion controller what is causing it. I have seen GTX users doing over 1.65V. So it must be undervoltage/overcurrent problem in my situation. I had similar problem with 7900GT and nothing fixed it. I hope this problem can be solved.

[ViperJohn]

Sorry, I could be more accurate.

It appears in most heaviest parts of 3DM06. (GT1 when door lifts in the middle of test, or in GT2)

Vgpu is 0V after that, so it must be that primarion controller what is causing it. I have seen GTX users doing over 1.65V. So it must be undervoltage/overcurrent problem in my situation. I had similar problem with 7900GT and nothing fixed it. I hope this problem can be solved.

Okay GPU loaded so the Vcore PS is probably shutting down on Vcore over current. I would bet that is related to the GTS only having one PCIe aux power connector and the weaker 3 phase Vcore PS used on the GTS. NV probably has tighter overcurrent settings on the GTS to protect the lower capacity card power input circuits and the lower phase count Vcore PS.

1.65Vcore on a GTX would need a hell of a lot of core cooling like cascade phase change. Can't even imagine how much loaded Vcore droop there would be as the GTX regulator allows a ton of loaded droop from a 1.450 desktop Vcore as it is.

Viper

[K404]

Slight diversion.

My card appears to have no idle-load voltage drop/droop, and is 1.35V, unmodded. Why the differences in core voltages between cards? My measure point is the same as posted on post 8 (used left-hand point)

If OVP is kicking in at 1.43V, thats only around 7% above standard (taking 1.35V as standard) which seems a bit close to the limit??

@SF3D..what core/shader clocks were/are you running? Have you seen anyone benching higher MHz than you?

[ViperJohn]

Slight diversion.

My card appears to have no idle-load voltage drop/droop, and is 1.35V, unmodded. Why the differences in core voltages between cards? My measure point is the same as posted on post 8 (used left-hand point)

If OVP is kicking in at 1.43V, thats only around 7% above standard (taking 1.35V as standard) which seems a bit close to the limit??

@SF3D..what core/shader clocks were/are you running? Have you seen anyone benching higher MHz than you?

Interesting. The GTX's stock Vcore is about 1.290.

Viper

[SF3D]

Slight diversion.

My card appears to have no idle-load voltage drop/droop, and is 1.35V, unmodded. Why the differences in core voltages between cards? My measure point is the same as posted on post 8 (used left-hand point)

If OVP is kicking in at 1.43V, thats only around 7% above standard (taking 1.35V as standard) which seems a bit close to the limit??

@SF3D..what core/shader clocks were/are you running? Have you seen anyone benching higher MHz than you?

This is weird... My unmodded voltage is 1.26V idle and 1.21V load

Core clock was 729MHz (real) and shader clock was 1944MHz. I think my core clocks are not highest and I will try to reach higher clocks today.

New 97.28 driver is best for me.

[K404]

is is possible you are running into a standard stability issue and the VCore isnt high enough, or not cold enough for the clocks?

Not tried the voltmods yet, probably wont until everyones happy with the voltmod points

[SF3D]

If vcore isn't high enough i'm getting lockups. Have to test it more now.

[syne_24]

Hey, what is the highest you can go on the core before needing to vmods witht the GTX? I just wanna make sure that is what holding me back and not some other issues. Right now stuck at 700 for some reason. My GTX is under water, temps are great, so it shouldnt be a problem. Thanks

[K404]

Another point... a 5K resistor for the GDDR puts the RAM to 2.04V, so to play safe, i`d recommend 10 or 20K instead

Still to try VCore, hoping for more details on black screens at 1.43VCore...core MHz too high, or some protection feature kicking in

Kenny

[SF3D]

Where did you get that idea?

5K resistor gives you 1.86V It's in the first post.

I did some tests today and reached 70k in 3DM01 and 12,6K in 06.
My GTS can do 743 ( real 756MHz) with 1,41V