[gmx-users] Question about GPU acceleration

[Jernej Zidar]

Hi,
  Did you try "playing" around with the number of MPI, thread tasks. I
was actually able to significantly improve the performance.

  Try for example overloading both GPUs like so:
gmx mdrun -v deffnm foo -ntmpi 4 -ntomp 2 -gpu_id 0011
or
gmx mdrun -v deffnm foo -ntmpi 2 -ntomp 4 -gpu_id 01

  Adding "-pin on" and maybe "-npme 0" should help as well.

  You can read more about the performance:
- http://www.gromacs.org/Documentation/Performance_checklist
- http://www.gromacs.org/Documentation/Acceleration_and_parallelization
(this one you might find particularly useful)

Cheers,
Jernej

On Fri, Nov 28, 2014 at 1:39 PM,
<gromacs.org_gmx-users-request at maillist.sys.kth.se> wrote:
> Message: 4
> Date: Fri, 28 Nov 2014 10:37:45 +0900
> From: "chip" <chip at bio.gnu.ac.kr>
> To: <gromacs.org_gmx-users at maillist.sys.kth.se>
> Subject: [gmx-users] Question about GPU acceleration
> Message-ID: <000801d00aab$e977eae0$bc67c0a0$@bio.gnu.ac.kr>
> Content-Type: text/plain;       charset="us-ascii"
>
> Greetings,
>
>
>
> I did simulations using 1 GPU and 2 GPUs. (System size : 203,009 atoms)