You need to set the environment variable CUDA_VISIBLE_DEVICES. I think the web page has information (see the page on gpus at ambermd.org).
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
On May 9, 2012, at 8:47 AM, Albert <mailmd2011.gmail.com> wrote:
> hello:
>
> I've got two GTX590 at a workstation and I am trying to run two
> series CUDA jobs in two different folder. I found that one is with 20
> ns/day and the left is only 6 ns/day. The system is the same. I am
> curious that the second probably is use the same GPU with the first one
> so that the efficiency is so slow.
>
> So, I am wondering, is it possible to specify GPU in pmemd.CUDA
> command so that two jobs won't use the same GPU for computation? I am
> using SUSE LINUX OS.
>
> thank you very much
> Albert
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 09 2012 - 07:00:02 PDT