Page 1 of 1

Mixing CPU and GPU clients?

Posted: Fri Apr 03, 2009 11:46 pm
by wjg
How well can simultaneous clients be run for the CPU and GPU versions? Right now I have two CPU clients for more cores, and I have a new compatible video card now. Can I run all three simultaneously?

Re: Mixing CPU and GPU clients?

Posted: Fri Apr 03, 2009 11:53 pm
by JimboPalmer
Yes, I am using 32 bit Vista and my HD2600Pro only takes about 4% of CPU time when folding, so on my dual core CPU, I run two CPU clients on low priority and one GPU client on slightly higher priority. Other OSs and/or faster Video cards would change your CPU time for the GPU client.

Re: Mixing CPU and GPU clients?

Posted: Sat Apr 04, 2009 1:02 am
by 7im
On what CPU type and speed? It does make a difference.

Re: Mixing CPU and GPU clients?

Posted: Tue Apr 07, 2009 8:53 pm
by 4ndr3w
i would think there would be SOMETHING that the cpu does better in these calculations that the gpu could offload to cpu. I figure that could possibly speed up the process.
have the cpu does something it does better, and have the gpu do what it does better. the result could be folding 1.5x to 2x faster!

of course, i have no idea what i'm talking about. the gpu probably does ALL this stuff better.

Re: Mixing CPU and GPU clients?

Posted: Tue Apr 07, 2009 11:17 pm
by bruce
4ndr3w wrote:i would think there would be SOMETHING that the cpu does better in these calculations that the gpu could offload to cpu. I figure that could possibly speed up the process.
have the cpu does something it does better, and have the gpu do what it does better. the result could be folding 1.5x to 2x faster!

of course, i have no idea what i'm talking about. the gpu probably does ALL this stuff better.
Exactly, except that your 1.5x to 2x is a huge underestimation of the capabilities of modern GPUs.

The GPU client does off-load the CPU roughly 98%. It (almost) uses only the GPU and is much, much faster. The science code had to be completely rewritten to take advantage of the large number of parallel processors found in a GPU, but that has only been done for one version of the science code. For other types of scientific analysis, the CPU code is all that's available.

Since the GPU code leaves the CPU almost idle, a second client can be installed to make use of the CPU and to perform some of the other types of analysis concurrently.