Overclock for GPU F@H

[Kougar]

I overclocked from 575/1188/1800 to 600/1500/1800 and went from ~2100PPD to ~2500PPD according to Fahmon. Doesn't compare well to GTX 280, but I'm happy with the boost!

I've never had reason to OC the shaders higher than 1500 since the GPU core is what boosts game performance, so I will have to test the GPU before I push them higher. Will also try underclocking the core and seeing how that impacts performance... my theory is an underclocked core will give the shaders a bit more power to work with. When I get a chance I'll test how the CPU overclock affects performance as well, because the GPU client pegged my 3.2GHz C2Q at 100% for one core... anything for those extra few hundred PPD!

[privater]

Thanks to alextooter,he write a program which provides a simple way to estimate your PPD performance in F@H
here is the small program download link used to benchmark your GPU CUDA performance，like super pi work on CPU.
not support SLI，and if you want to use it with 177.35 driver in vista make sure delete nvcuda.dll in the directory.


Because I don't have enough permission to upload images and ur-l,so please change the "*" to "." to download this program

and also you can download it in mediafire:

xp version：www*mediafire*com/?mbogfnqu95z
vista version : www*mediafire*com/?mmwbtnrgwch

run this program and wait,it will open a cmd window and display infomation such like:

Code: Select all

--------------------------------------------------------------
There is 1 device supporting CUDA

Device 0: "Quadro FX 570M"
  Major revision number:                         1
  Minor revision number:                         1
  Total amount of global memory:             268107776 bytes
  Number of multiprocessors:               4
  Number of cores:                                 32
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       16384 bytes
  Total number of registers available per block: 8192
  Warp size:                                     32
  Maximum number of threads per block:           512
  Maximum sizes of each dimension of a block:    512 x 512 x 64
  Maximum sizes of each dimension of a grid:     65535 x 65535 x 1
  Maximum memory pitch:                          262144 bytes
  Texture alignment:                             256 bytes
  Clock rate:                                    0.95 GHz
  Concurrent copy and execution:                 Yes
Using device 0: Quadro FX 570M
Processing time: 14.922157 s

Using device 0: Quadro FX 570M
Processing time: 20.002654 s

Using device 0: Quadro FX 570M
Processing time: 18.668309 (s)



Total time is 53.593121 s

Press ENTER to exit...


Then you can set different core/shader/mem freqs to see whats the difference in cuda ：）

[zman977]

Good call on this one guys. I was able to unlink my shader and core frequency on my 8800 GTS's, so I lowered the core clock to 385mhz, lowered the memory clock to 595, and raised the shader clock to 1513 so far. My performance in the app has gone up about 30 percent, but my temperature for the card has dropped 10 degrees C so far. HUGE difference since my G80 core cards run hot to begin with.

so wait if i read your post correctly, you lowed both your core clock and mem clock but raised your shader clock? and this gave you a 30% performance increase?! i might do this to mine because i also have 2 8800 (320meg) gts in sli and they are around 78C when folding. But i do wonder how this will affect my gameing performance? Also did you do this with a moded firmware or with the nvidia n-tune software?

[privater]

here is the Result from my 8800GT
There is 1 device supporting CUDA
Core/Shader/Mem:

615/1650/2000 (Temp:60c)
Total time is 13.751111 s

675/1950/2000 (Temp:61c)
Total time is 12. 842419s

741/1950/2000 (Crash when Fah fullload @ 5 Minutes ,Temp :64c)
Total time is 12.721728 s

675/2050/2000 (Stable for 20 hours,Temp :61c)
Total time is 12.575684 s

741/2050/2000 (Crash when load @ 20 seconds,Temp:XX)
Total time is 12.493259 s
according to these result,I find my 8800GT works in best performance temprature balance at 675/2050/2000.
my card use original cooler from EVGA and added an Intel cooler fan on it,set in a case which can provide good air flow.
If I upgrade my cooler,the freqs limit might push farther.

[hermes3]

so here is mine with GF 9600GT 512MB (700/1750/1900MHz) - drivers 175.16

Code: Select all

There is 1 device supporting CUDA

Device 0: "GeForce 9600 GT"
  Major revision number:                         1
  Minor revision number:                         1
  Total amount of global memory:                 536543232 bytes
  Number of multiprocessors:                     8
  Number of cores:                               64
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       16384 bytes
  Total number of registers available per block: 8192
  Warp size:                                     32
  Maximum number of threads per block:           512
  Maximum sizes of each dimension of a block:    512 x 512 x 64
  Maximum sizes of each dimension of a grid:     65535 x 65535 x 1
  Maximum memory pitch:                          262144 bytes
  Texture alignment:                             256 bytes
  Clock rate:                                    1.75 GHz
  Concurrent copy and execution:                 Yes
Using device 0: GeForce 9600 GT
Processing time: 4.106722 s

Using device 0: GeForce 9600 GT
Processing time: 14.537085 s

Using device 0: GeForce 9600 GT
Processing time: 6.618505 (s)



Total time is 25.262312 s

Press ENTER to exit...

[mentalinc]

Your GPU doesn't support CUDA or you need to update your GPU driver..

um yeah it does....
8800GT
drivers: 177.35

[eliot1785]

So far using a 8600M GT plus a C2D 2.5GHz and 800MHz FSB, I've gotten the best possible performance out of the following overclock settings (using RivaTuner):

GPU Core: 475 MHz
GPU Mem: 400 MHz
GPU Shader: 1250MHz

These are actually the defaults except that the GPU Shader defaults to 950MHz, so I increased it by 300MHz.

Using these settings, I complete 1% of a WU in about 67-68 seconds. nVidia System Monitor says my GPU temperature is up to 82-83C though (is this too high?). i8kfanGUI puts it at 76C but I'm guessing nVidia will have a more accurate measurement of its own cards.

Overclocking any component further doesn't seem to have any effect on the client speed, but I think that is partly because things are CPU-bound. If the client were improved to put less of a load on the CPU, I might be able to get more performance by overclocking further.

[Mr. Scary]

just found this thread, here's where i'm at x2. (second card is within ten seconds of these stats)







these temps are with the fans at 65%

did anyone really determine if moving the core & mem clocks down or even below stock increases ppd? or keeps the same and decreases heat.

M$

[MtM]

toTow, since the thread is bumped already I just read your statement and I'm confused. In the ppd thread we did establish that your only reason you beat my older submission was because you were able to have shaders at 2050 or something and much lower core speeds. I did 750-1800 then ( core/shaders ) as that was a nice balance between shaders and game performance ( core clock ). So either I am mistaking you with someone else, you should have experienced the influence of shader clocks on 9600gt. I dropped core 25mhz but raised shaders to 1900 which is last stable point atm with stock cooling/volts. 1925 eue's sometimes, 1950 within 3 frames 100%.

[extrasalty]

My MSI 8800GT (stock 660/1650-around 4900 PPD) was stable at 720/1728. After I flashed my BIOS with 1.1v now I'm at 756/1782- I know I could have gone further- F@H is a little less picky than ATI tool and Furmark, but I play a lot and these frequencies are rock solid. On average it does 1:20s (5184 PPD)- so it is around 5% increase in PPD compared to stock and the shaders are overclocked by about the same amount. Memory overclocking just makes it unstable without any benefit in games or F@H. As for the temperatures- the card has heat pipe cooler and keeps it below 60 C under load- if I keep the fan at the same speed, it idles at around 45 C; the PCB temperatures move around 10-12 C . The voltage increase did raise the temperature by 1-2 degrees C. I've noticed that playing with frequencies doesn't really change the heat output- voltages is what matters. So I'm pretty sure my chip will not go higher unless I raise the voltage further, which I'm not comfortable with.
One thing I noticed on Nvidia cards is that the PCB warps downwards with time- the cooler weight and the temperature must be the reason. I'm pretty sure that similar conditions caused the RRoD on Xbox and I have the card supported by a BIC pen- better safe than sorry.

[MtM]

On another note..

I can game/furrmark/ati tool 100mhz higher on shaders then I can fold ( though I expect the exact limit varies from client to client and from wu to wu ).

From your cooling sollution it sounds like your card comes stock with an aftermarker cooler and there your clocks don't influence your temps as much. For normal situations clocks = temps. Only if your cooling solution is up for the task it will move to clocks != temps but that happens only when you already have surpluss of cooling power.