Unless you are doing some VERY particular calculations, there is almost
no gain to be had from using more than one GPU.
Replica exchange for instance is one of the situations where you use
more than one GPU.
For SMD, I cannot imagine a reason to use multiple gpus.
Try pmemd.cuda without MPI and it will work, and it will almost the same
speed.
BTW: the same applies to regular MD. Using pmemd.cuda.MPI for regular MD
has very little gain. You are better off running 2 simulations with
different seeds, one in each gpu.
adrian
On 8/16/17 3:16 PM, Xing wrote:
> Dear all,
>
> I'm trying to use pmemd.cuda.MPI to run Steered MD (jar=1) on two GPUs. However the system will blow up within 10 steps while the pmemd.MPI will run without a problem on CPUs. I used pmemd.cuda.MPI
> for "normal" molecular dynamics and it worked fine. Could this be some bug caused by the GPU precision limit? Is there a workaround if it is a bug?
>
>
> Thanks and best wishes,
>
> Xing
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
Dr. Adrian E. Roitberg
University of Florida Research Foundation Professor
Department of Chemistry
University of Florida
roitberg.ufl.edu
352-392-6972
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 16 2017 - 15:00:02 PDT
