Hi Florian
Before we go further, have you checked regular MD with 2 GPUS? Does that
work ok in your setup ?
Thanks
adrian
On 2/15/17 7:50 AM, Hofer, Florian wrote:
> Dear AMBER Developers,
> we are using the constant pH calculation algorithm in AMBER, espacially the newly implemented version for pmemd.CUDA in AMBER 16. Calculations of the same system using a single GPU and the pmemd.cuda executable work fine, however when we were trying to do the calcualtion with 2 GPUs and using the pmemd.cuda.MPI executable we noticed that the calculations get halted at about 15000 steps.
> The caluclations produce regular outputfiles until they stop and the jobs drop from the GPU, without any error messages. However the jobs remain active on the CPU with 100% each.
>
> The caluclations are done on a GPU cluster with the following specs:
>
> MPI program: mpich-3.2
> AMBER version: AMBER 16 updated on 10 Feb 2017
> Queue system: SGE-6.2u5p3
>
> Hardware:
> Peer to Peer: Enabled
> GeForce GTX 780 2x or GeForce GTX TITAN X 2x
> Intel(R) Xeon(R) CPU E5-2609 v2 2x
>
> Has anyone else noted similar behavior or can help us figure out the problem?
>
> Really appreciating your help and best wishes,
> Florian Hofer
> _______________________________________________
> 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 Feb 15 2017 - 06:30:03 PST
