Hardware Canucks

Hardware Canucks (http://www.hardwarecanucks.com/forum/)
-   HardwareCanucks F@H Team (http://www.hardwarecanucks.com/forum/hardwarecanucks-f-h-team/)
-   -   Folding@home gpu benchmarks (http://www.hardwarecanucks.com/forum/hardwarecanucks-f-h-team/11896-folding-home-gpu-benchmarks.html)

chrisk November 17, 2008 04:49 PM

Folding@home gpu benchmarks
I was looking through the thread that Sky had concerning what we wanted in GPU reviews, and we all felt that a fah review would be very useful for some of us. I agree that such a review would be pretty popular and useful, but I also agree that doing so has a few problems (like testing time, especially if you have to test for a bunch of games, etc)

I was wondering a few things;
1. Any good sites that you use to find folding@home GPU reviews, similar to something like this: PCGH - Review/Benchmark: Folding@Home on Nvidia GPUs: PCGH benchmarks

2. How hard/viable/likely would it be for someone here at HWC to do something like this and post to a sticky or something? I don't think I have the expertise (or resources) to do such a thing myself, but such a post I think would get lots of attention from the folders around the web, and perhaps even help recruit more folders for our team.

What do you guys think? Or has this been considered already? Old news to the experienced folders? Links to great sites that do this?

Eldonko November 17, 2008 05:04 PM

This guy's work is my favorite:

Folding Forum • View topic - NVIDIA GPU2 PPD Thread [Preliminary Results]

Alwaysrun November 17, 2008 05:46 PM

Nice comprehensive list he has there but I wish there was another one for ATI cards. About all I ever see is ATI sucks for folding. Sad but true, and waiting for a new GPU client or something from ATI to help close the gap is like waiting for the second coming. Didn't ATI sort of start using GPUs to fold first? And exactly why do Radeon users get such low ppd? Is it got to do with the shader core?

frontier204 November 18, 2008 06:38 PM


Originally Posted by Alwaysrun (Post 113747)
Nice comprehensive list he has there but I wish there was another one for ATI cards.

There is an ATI benchmark table as well, as a sticky in the "ATI specific issues" thread of the Folding Forums. The post is called "Optimal CPU - GPU comparison table [moderate to larger WUs]". (I can't post the link because foldingforum.org has been unreachable for me for a week or so.)


Originally Posted by Alwaysrun (Post 113747)
Didn't ATI sort of start using GPUs to fold first?

Yep, first with the X1000 cards, then the 2-3XXX cards.

I know when I started GPU folding, my 4850 actually performed WORSE than most 3870 cards in terms of PPD (~1200-1500) because the core only used 320 of the 800 stream processing units on my card. The other issue was very high CPU usage on the core. Now, the CPU usage has dropped, but every WU I attempt gives me UNSTABLE_MACHINE...
Some of the older posts you see about nVidia >> ATI in folding is from the fact that when nVidia was in its initial stages, the PPDs were through the roof for some reason.

Eldonko November 19, 2008 08:57 AM

I cant fold at all with my 4670s, sucks I have to use SMP. :/

chrisk November 19, 2008 09:28 AM

Yeah I picked up a 260 a week or so ago to replace my 4870 and have been on folding fire ever since

SKYMTL November 19, 2008 10:22 AM

I would love to add this to my benchmarks but it seems like the PPD really fluctuates based on the project and driver revision.

Eldonko November 19, 2008 10:27 AM

You can test the same WU by adding verbosity 9 flag and deleting work and queue until you get the same one. PITA I know, but possible. You could also set a certain point in the WU to take the PPD, say between frames 30 and 40 for example. Driver revision has an effect on everything, not just PPD.

SKYMTL November 19, 2008 11:19 AM

I know but it has already been brought up: PPD testing takes alot of time.

Once everything calms down (yeah right) I will probably do a GPU Folding article.

Eldonko November 19, 2008 11:29 AM

Understood, I just wanted to point out that it is possible to test in a replicable format which uses proper methodology. PITA and time consuming though.

All times are GMT -7. The time now is 01:55 AM.