Folding@home gpu benchmarks

[sswilson]

I'd also be a bit concerned about the effects of CPU on gpu folding. The PPD for my 8800GTS varies by up to 1000 PPD depending on what rosetta is doing with the other core, even though that core never goes above 70%. Not sure if that would be in evidence on a clean system with nothing else running, but it does make me wonder.

[chrisk]

I understand the problems doing this benchmark Sky so no problems there. I was mostly wondering if somebody out there was doing this or if somebody in our community could do something like this and post in the forums via sticky, etc. I don't have the time but mostly resources ($$$) to do so, but I believe that there is a great opportunity for a reliable benchmarking site to gain readership. Perhaps somebody smarter than I can create a benchmarking utility that simulates fah crunching that would make this easier.

[chrisk]

Originally Posted by sswilson

I'd also be a bit concerned about the effects of CPU on gpu folding. The PPD for my 8800GTS varies by up to 1000 PPD depending on what rosetta is doing with the other core, even though that core never goes above 70%. Not sure if that would be in evidence on a clean system with nothing else running, but it does make me wonder.

I agree that the cpu is a variable as well. In gaming benches, the system platform, OS, GPU, and drivers are all variables. In fah, if I understand correctly, the core and project being sent down adds two additional variables.

[1Tanker]

Originally Posted by chriskwarren

I understand the problems doing this benchmark Sky so no problems there. I was mostly wondering if somebody out there was doing this or if somebody in our community could do something like this and post in the forums via sticky, etc. I don't have the time but mostly resources ($$$) to do so, but I believe that there is a great opportunity for a reliable benchmarking site to gain readership. Perhaps somebody smarter than I can create a benchmarking utility that simulates fah crunching that would make this easier.

The Tech Report has been doing it for a couple years now(maybe a bit less)..but their results are extremely low. They use what i would say is a "script".. a CD that one of their forum members made.. to bench F@H. Actually, come to think of it.. it was for CPU's ... not GPUs. Anyways..it would be interesting to do/see. I think as others have mentioned, that if you DL a WU for ATI and one for nVidia..save them..then when you run it.. make sure to unplug from the net. Just delete the "work" folder after each run..then you're all set for the next. I don't think that Stanford would be too happy about people DLing WUs as benchmarks though, but you wouldn't have to do it very often.

[Eldonko]

All you have to do is not run any other process when benching ppd for it to be consistent. I mean if you run any benchmark you arent going to run rosetta or similar at the same time.

[Alwaysrun]

Well I've been getting 3094ppd average according to FAHMON with my Palit 4850 at stock (it dosn't like my normal gaming clock of 670/1023) using the CPU client as well. So I can attest that CPU folding seems to be effecting the GPU because without the CPU client working I get 3298 with the 4850 running solo. The law of diminishing returns is really giving me a mule kick to the cranium here as I have to run the CPU full time just for an extra 176ppd. Well thats not exactly true as sometimes I get nice projects like 2613 that gives 697ppd so it makes using the CPU worth it a few times a week. I can really see the problems involved with so many variables and I think now that posting ppd averages in reviews is just to time consuming for lets face it rather soft numbers.

Does anyone know what sort of "Average" that review unit SKYMTL put up on the buy/sell yesterday would get? I'm sorta eyeballing that lil gem.
http://www.hardwarecanucks.com/forum...x-maxcore.html

[Eldonko]

Probably 6500-7000 from a 260 with a shader OC. The card with the best PPD/$ ratio is the 9800GX2.

Your PPD goes down when you run CPU + GPU client because you have an ATI card. ATI cards use way too much CPU power so your SMP clients is too slow to make it worth it. For ATI it is best to use GPU client alone, for Nvidia GPU2 only uses about 10% of CPU so you can use both.

[Alwaysrun]

Ahh I see. Thanks Eldonko.

BTW I'm just using the x86-systray. I tried like hell to get Deino SMP going but I'm still a folding nub and I got halted at the password part and it wouldn't work. Really I shouldn't be mucking with it, I know because my lowely e8400 has only 2 cores and as you said my ATI GPU is more processor dependant. Nonetheless I wanted to tinker and check out something new.

Sorry for hijacking your thread chris so to jump back to it I think your idea is great one and I'd love to see a thread started with ppd averages that users could submit. I saw a thread started along those lines a couple days ago but I can't for the life of me find it now.

[SKYMTL]

Originally Posted by Eldonko

Probably 6500-7000 from a 260 with a shader OC. The card with the best PPD/$ ratio is the 9800GX2.

Don't forget to add to that the fact that the GTX MAXCORE I have for sale has 216SPs versus the standard card's 192.

[chrisk]

Originally Posted by Eldonko

Probably 6500-7000 from a 260 with a shader OC. The card with the best PPD/$ ratio is the 9800GX2.

Your PPD goes down when you run CPU + GPU client because you have an ATI card. ATI cards use way too much CPU power so your SMP clients is too slow to make it worth it. For ATI it is best to use GPU client alone, for Nvidia GPU2 only uses about 10% of CPU so you can use both.

I'm getting 7100-7200ppd on my Evga 260 core 216, My shaders are OC right now to 1550mhz, and the gpu temps get to around 52-58c. Still haven't pushed the shaders until I get a eue yet (I actually haven't got an eue since I moved from an Ati card to nVidia)