Folding on nVidia

Interesting. IIRC, the reason they didn't get F@H running on nVidia was because of the branch-prediction of the ATi cards as well as some technical detail of the FP units that I don't recall. I see that he mentions that nVidia will be releasing 64-bit cards later this year. That could certainly help their chances for F@H.
 
Nope. Nvidia cards earlier than G80 can't support our looping constructs fully. We are having what we have initially tracked to issues with drivers/compilers currently with G80 which are preventing the FAH client from working correctly. Once we get past this, we shall see if everything checks out, and then start looking at performance.

CUDA is neat, but is single vendor only and would require a complete code rewrite. We'd like to stay with DX9 for awhile longer, and then look into other options for future cores, likely starting with DX10.
 
Well, while we're on this subject: For the current ATi FAH client, how much of an increase in performance is being seen over CPUs such as the Intel E6300s (with both cores running FAH clients).

Also, same question with regards to the PS3 FAH client.
 
Any news about folding on the new ATI R600 GPU that is set to go on sale to the public in March?

It will be released soon... within about 5 weeks.

Thanks
 
kind of a thread jack but anyone know anything about the PS3 client??? Ihave 2 that are idle just waiting to fold
 
CoW]8(0);1030654229 said:
Well, while we're on this subject: For the current ATi FAH client, how much of an increase in performance is being seen over CPUs such as the Intel E6300s (with both cores running FAH clients).

Also, same question with regards to the PS3 FAH client.
Click to expand...

The CPU and GPU/Cell clients are running different cores. The GPU/Cell cannot easily/efficiently do some of the cores the CPU is doing. Conversely, the GPU/Cell cores are incredibly math heavy and are largely impractical for the CPU until the CPUs have lots of cores. Apples to apples, i.e the CPU running the GPU cores and heavily tuned for x86 SSE, the GPU is 20-40X a single core of a Core2. (it depends on how MAD heavy vs other ops the particular kernel is).

As for PS3, until it's public, we can't really comment. You can dig up some of Vijay's comment on PS3 in the forums to get a hint at performance. The biggest improvement for PS3 is that the client should be much easier to use as you don't have issues with different driver revisions to deal with, and it won't interfere with your normal PC usage (obviously).
 
You must log in or register to reply here.
Back
Top