GPU folding on nvdia based boards?

[Freak'n Big Panda]

Clearly it would be possible to implement on G80 based cores, maybe not G70. I'm not too sure of what flexibility is needed in the shaders in order to implement the client (maybe somebody could clarify this for me?). Regardless are there any plans in the works to get the FAH client running on nvidia boards? Or is it going to be ATI only for the foreseeable future?

[Tim Murray]

The official story is that there's a driver bug preventing FAH from working properly on NV cards. NV has acknowledged the bug but has not issued a fix yet. Once that happens, development will continue, and who knows if they'll find another bug or something.

[BRiT]

One would think it would become a priority if Nvidia was serious about their entry into the GPU-CPU realm. They released a couple of Press Releases stating they were capable of performing intensive mathematical tasks, but they have nothing to show for it. Not even a Folding@Home client.

Sounds like more lip-service from NVidia like their monthly driver updates... :roll:

[Jawed]

I think NVidia is focussed as follows:

Linux-based researchers in universities
software houses developing a "workstation" package to niche high-performance industries, such as the oft-quoted seismic and option pricing solutions
end-user companies that can afford to do their own in-house R&D on blue-sky computingThe GPGPU universe is bigger than Brook and F@H. That combination has proven the concept, but this is a fast-moving industry. People know GPGPU is good stuff, they just want to run their own applications (or at least prove it's worthwhile for them).

Jawed

[BRiT]

Though what better way to show off your GPGPU strengths than with some simple / proof-of-concept such as F@H?

[Freak'n Big Panda]

I guess it's pretty simple from a financial standpoint, I really doubt there's much money in getting FAH running on your hardware, obviously it would cost a bit to rework the drivers in order to make it possible. A quick question though, can FAH technically run on any SM3 hardware? Would NV30/G70 be able to pull it off? Or do the requirements go beyond SM3 specs?

[Tim Murray]

Though what better way to show off your GPGPU strengths than with some simple / proof-of-concept such as F@H?
The question is, would they win against R600? Based on the teraflop-in-a-box demo, I'd think "hell no." But for that matter, would they win against R580?

NVIDIA doesn't care about GPGPU as a whole--they care about CUDA. If FAH comes out with a CUDA version, and they say things like "it runs faster/means we don't have to worry about drivers/blah blah blah," NVIDIA will support it. Otherwise, it's about as low-priority for them as possible.

[Bubonic]

The question is, would they win against R600? Based on the teraflop-in-a-box demo, I'd think "hell no." But for that matter, would they win against R580?

NVIDIA doesn't care about GPGPU as a whole--they care about CUDA. If FAH comes out with a CUDA version, and they say things like "it runs faster/means we don't have to worry about drivers/blah blah blah," NVIDIA will support it. Otherwise, it's about as low-priority for them as possible.

If I remember right, the gflop difference between R580 and G80 is small. Both being 300+gflops. I do qustion G80's GPGPU abilities, becuase all I have seen is paper spec.