dslreports logo
 
    All Forums Hot Topics Gallery
spc
Search similar:


uniqs
647

mr sean
Professional Infidel

join:2001-04-03
N. Absentia

1 recommendation

mr sean

Folding@Home Hardware Question

Greetings fellow folders!

I'm curious about how various discrete GPUs perform running the current client and am hoping some of you might share your impressions and include a general idea of Points Per Day (PPD) for your particular GPU.

It seems the nVidia cards, especially those with CUDA cores, really perform well, but it would be interesting to see a comparison with the ATI/Radeon cards. I'm trying to find a sweet spot between price and performance for a new boxen build using multiple (2-3) graphics cards (I'm leaning towards the GeForce 660Ti).
Thanks in advance.
Scilicet (banned)
Spaced Out
join:2005-04-11
Aurora, CO

Scilicet (banned)

Member

I run a GTS 450. Unfortunately, there was a marked decrease in video performance and speed. The client had to be removed. In my opinion, the best increase in PPD was noticed when I replaced the dual core CPU with a 6 core.

mr sean
Professional Infidel

join:2001-04-03
N. Absentia

mr sean

said by Scilicet:

In my opinion, the best increase in PPD was noticed when I replaced the dual core CPU with a 6 core.

Which processor?
I agree that upgrading a processor usually offers a big boost. I'm running an Intel SB-E 3930k with a EVGA GTX 680+ in one boxen, quite happy with that performance. The discrete GPU runs its own client along with the processor easily in the 7.x.x version, and the 7.3.6 version seems especially optimized for CUDA cores.
What I'm really looking for is a means of getting the most out of a single build for a dedicated folding rig. After the initial choice of processor the next line of improvement will probably come with nVidia CUDA based GPUs. The 660Ti has a nice price point, is very similar to the 670 in spec, and might offer the best price/performance option for multiple cards.
Price-wise the 660Ti is about half the cost of a 680, but two 660s might offer 40-60% more production than a single 680. This might also offer an option for other folks when considering upgrade options outside of a new processor/mobo. Improve your gaming, streaming, etc and improve your folding production too.

howieh
4 more weeks
MVM
join:2000-08-06
Metuchen, NJ

howieh to mr sean

MVM

to mr sean
Pin See Profile recently posted some info on upcoming GPU changes:
»[F@H] GPU QRB is getting closer!
Scilicet (banned)
Spaced Out
join:2005-04-11
Aurora, CO

Scilicet (banned) to mr sean

Member

to mr sean
Click for full size
I run an AMD Phenom II x6 1045T overclocked at 10% on an ASUS Crosshair V formula mobo. As you can see from the team stats above I have a fairly low WU count for the amount of points I have accumulated. My Zotac GTS 450 GPU is a low end model ~$90. Probably the cause of low video performance when running a F@H client.

mr sean
Professional Infidel

join:2001-04-03
N. Absentia

1 recommendation

mr sean to howieh

to howieh
Ah...so it's the OpenCL libraries and not the CUDA in this client version. Thanks Howie! Now I'll have to work twice as hard to figure this out.
said by Scilicet :
As you can see from the team stats above I have a fairly low WU count for the amount of points I have accumulated.
That's a nice return on the points. I think I'm averaging about between 5k-6k per WU according to Kakao Stats, but my WU count is askew thanks to old school tallying from 2001-2005.

So what kinda numbers are the newer 7xxx Radeon cards generating?

Sunny
Runs from Clowns

join:2001-08-19

Sunny to mr sean

to mr sean
Sure is nice to see you showing a renewed interest in Folding@Home.

Big welcome back.

mr sean
Professional Infidel

join:2001-04-03
N. Absentia

1 recommendation

mr sean

I just missed hanging around you guys...

Pin
Premium Member
join:2002-06-19
The Dungeon

1 edit

Pin to mr sean

Premium Member

to mr sean
said by mr sean:

Ah...so it's the OpenCL libraries and not the CUDA in this client version.

That version is not quite out of beta just yet, but it looks like going forward that OpenCL is the development path. That could change at any moment.

That being said, there is a Cuda flip flop with OpenCL. Currently, AMD ( in production ) uses 1 CPU and Nvidia ( in production ) uses 0 CPU. Here comes OpenCL. With the OpenCL core 17, AMD uses 0 CPU and NVidia uses 1 CPU.

So it all depends on how you want to use the rest of your resources.

But wait, there's more:
»folding.typepad.com/news ··· -17.html

Looks like they might have already double the processing speed of OpenCL cores before you ever got a chance to peek at the first OpenCL core.

Edit:
Here a little peak at some of the core data collected during testing, before the NVidia Titan came out.

»docs.google.com/spreadsh ··· sh#gid=4

mr sean
Professional Infidel

join:2001-04-03
N. Absentia

mr sean

Ooooo....spreadsheet!
That will give me something to fun to do instead of the stack of peer reviewed articles that seems to grow hourly.
I'm guessing those are some of the newer AMD drivers? I'll have to look at Radeon a bit more closely. But those are some exceptional production leaps on Vijay's blog.

Thanks!