Which GPU Benchmarks on Tom's Hardware GFX reflect F@Hgpu

[der foos]

Which benchmarks on Tom's Hardware Graphics Card comparison list best reflect Folding @ Home GPU performance?

THGC list is useful and a good upto date resource - trying to determine F@H impact by card.

#2
What current additional resources/data possibly available to overlay on the list to help isolate F@H performance?



thx - foos

 

[Mousemonkey]

There might have been something like that somewhere in the old F@H thread but I haven't seen anything like that recently.

 

[hunter315]

The issue is that there are multiple GPU cores for F@H, the nVidia one and the ATI one are very different so you would have to bench each card on its core individually and that would get very time consuming. Are you planning to select a GPU specifically for F@H?

 

[der foos]

There might have been something like that somewhere in the old F@H thread but I haven't seen anything like that recently.

hi Mouse

we agree - it is hard to compare the stats of modern cards in this area, so the question.

thx - foos

 

[amnotanoobie]

Unless there is something new done with the ATI implementation, the nvidia would still be a better performer for F@H. Last time I read something about F@H GPU, the nvidia implementation uses CUDA, whilst the ATI uses Brook (which isn't capable of using an ATI card to its fullest).

If the ATI implementation has been updated to use Stream/CAL, then it might actually start competing with the nvidia based one.

 

[der foos]

The issue is that there are multiple GPU cores for F@H, the nVidia one and the ATI one are very different so you would have to bench each card on its core individually and that would get very time consuming. Are you planning to select a GPU specifically for F@H?

hi Hunter

A summary would be helpful - we agree the solutions differ. In our view even with in Brand as well as Tech issue make for some confusion when comparing.

We are looking at F@Hgpu as a primary work load
- but the machines also do day to day activity (kids/work etc.) gaming is actually a bit lower down the scale.

Have some 9600 GTs today and trying to determine a path that might last for a few years - so thought we would ask ;-).

Thx - foos

 

[der foos]

Unless there is something new done with the ATI implementation, the nvidia would still be a better performer for F@H. Last time I read something about F@H GPU, the nvidia implementation uses CUDA, whilst the ATI uses Brook (which isn't capable of using an ATI card to its fullest).

If the ATI implementation has been updated to use Stream/CAL, then it might actually start competing with the nvidia based one.

hi not a noobie

Will research that aspect - know there a lot of folks using ATI/AMD with success.

Absolute performance is helpful and Nvidia has done well for us - but looking at several factors including cost position so scoring for us would look at
cost of upgrade for 4 machines to the relative delta of a better solutions. Nvidia may have a clear advantage - however have not looked at it.

To your point will dig in and see what the current position is.

thx - foos

 

[Mousemonkey]

As far as Nv cards go the more SP's or 'CUDA Cores' (gotta love that term) the better, the 8800GT 512 does quite well but the 9500GT struggles and I haven't bothered with the GT220 as it only has 12 more than the 9500.

 

[der foos]

As far as Nv cards go the more SP's or 'CUDA Cores' (gotta love that term) the better, the 8800GT 512 does quite well but the 9500GT struggles and I haven't bothered with the GT220 as it only has 12 more than the 9500.

Cuda Core per Dollar - CC$ ratio you say ;-)

Will take a minute but that sounds like a spreadsheet sort of thing we can work on - good suggestion.

thx - foos

 

[Mousemonkey]

For Nv cards more cores are better and in that respect regardless of it's gaming prowess Fermi is gonna [:mousemonkey:4] at folding.

 

[Mousemonkey]

Then there is the recent abomination from EVGA of course, I have wondered whether the two GPU's can both be used for folding or just the primary but as yet I have found very little info on the thing.

 

[cyberkuberiah]

Unless there is something new done with the ATI implementation, the nvidia would still be a better performer for F@H. Last time I read something about F@H GPU, the nvidia implementation uses CUDA, whilst the ATI uses Brook (which isn't capable of using an ATI card to its fullest).

If the ATI implementation has been updated to use Stream/CAL, then it might actually start competing with the nvidia based one.

yes , the ati client is still based on r600 (hd 3000) and they still havent updated it . perhaps the next major update would be not cuda or brook , but openCL to eliminate unnecessary development of different clients , just a consolidated one .

the folding team develops clients , but has to works with both ati or nvidia . this "bottleneck" would be reduced .

it would be interesting to see the folding performance on fermi with the supposed 900 gflops double precision and 1800 single precision , and radeon 5870 with 544 gflops double/2720 gflops single precision (yeah , ati has more single precision but lesser double precision performance , it really dont know much either are used in folding ) with a core to take full advantage .

 

[randomizer]

The GPU 3 client will use OpenCL and OpenMM which should allow the use of protected memory (something the NVIDIA client can already do) and significantly improve performance on ATI hardware. In the short term NVIDIA cards are a far better option though.

 

[kg4icg]

Currently a 5870 folds about as well as a 8800GS, the 295GTXs are the king of folding in the gpu arena until something changes.