Re: [AMBER] Extra Points support in pmemd.cuda.MPI

From: Jason Swails <jason.swails.gmail.com>
Date: Wed, 10 Apr 2013 08:27:41 -0500

On Apr 10, 2013, at 8:09 AM, Thomas Evangelidis <tevang3.gmail.com> wrote:

> It's on the to-do list AFAIK. Can you just run 2 separate jobs in the
>
>> meantime to utilize both GPUs? It never hurts to have 2+ simulations with
>> different velocities/seeds of the same system.
>>
>
> Actually I am running 4 simulations of the same system with different
> velocities :)
> Have you ever managed utilizing both GPUs of the same node by running two
> separate jobs? I tried that without any success.

Sure, I do it all the time. There are a couple strategies. The best one is to use mpirun or mpiexec if it supports MPMD (multiple-program multiple data). That depends on the specific MPI implementation you are using.

The other way is to launch 2 jobs in the background (end the command with a &) and then use the command 'wait' to make the script halt until both jobs are done.

HTH,
Jason

--
Jason Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Apr 10 2013 - 06:30:03 PDT
Custom Search