[gmx-users] Performance of GPU Workstation (GPU selection)

Carsten Kutzner ckutzne at gwdg.de
Thu Dec 4 19:13:19 CET 2014


Hi,

On 03 Dec 2014, at 22:12, mail at 2718-mail.de wrote:

> Dear gromacs Users,
> 
> We are currently considering to build a GPU Workstation for Gromacs Simulations.
> 
> Our typical Systems  consist of ~3000 Atoms(~30 per Molecule) solvated in 10000-40000 tip4p water molecules.
> 
> 
> We thought about getting a dual Processor node( E5-2630v3 or 2640 (Hasswell, dual 8 Core with 2.4 or 2.8Ghz)) and 4 GTX 780.
> 
> Anny remarks about this setup? What performance could be expected from such a system?
Are you planning to run a single copy of that MD system across the 4 GPUs? I am not
sure that this will scale well.

I have benchmarked a ~80000 atom test system on a dual processor E5-2670v2 (altogether
20 physical cores on a node @ 2.5 GHz) with 4x GTX 780Ti, which is slightly faster than
the GTX 780. Even with this larger system I get 21, 46, 57, 61, 64 ns/day using 
0, 1, 2, 3, 4 of the GPUs (at a 2 fs time step). So more than 2 GPUs does not really 
make sense here I would say.

Best,
  Carsten

> 
> Here I've uploaded an example input, if anyoneme with a similar system would be kind enough to run a short benchmark?
> 
> https://www.dropbox.com/sh/2na5wi530tb61sg/AACZavIW0FspgPAV3BkakXf0a?dl=0
> 
> 
> -- 
> Gromacs Users mailing list
> 
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
> 
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> 
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-request at gromacs.org.


--
Dr. Carsten Kutzner
Max Planck Institute for Biophysical Chemistry
Theoretical and Computational Biophysics
Am Fassberg 11, 37077 Goettingen, Germany
Tel. +49-551-2012313, Fax: +49-551-2012302
http://www.mpibpc.mpg.de/grubmueller/kutzner
http://www.mpibpc.mpg.de/grubmueller/sppexa



More information about the gromacs.org_gmx-users mailing list