[gmx-developers] Gromacs 4.6 on an inhomogeneous cluster

Szilárd Páll szilard.pall at cbr.su.se
Tue Oct 30 19:25:29 CET 2012

On Mon, Oct 29, 2012 at 2:51 PM, Berk Hess <hess at kth.se> wrote:

> Hi,
> Compiling in GPU functionality has no effect on performance.
> We have never timed the effect of compiling in MPI and OpenMP,
> but I would expect it to be very small, probably less than 1%.
> But it would be good to try that once on a small system.

I've just tried (again) the performance difference in thread-MPI-only runs
using mdrun compiled with vs without is negligible (<0.5%).


> Note that pure OpenMP can be much faster than MPI.
> The difference in performance between SSE2 and SSE4.1 is minor,
> less than 5% (AVX is a lot faster though).
> So I think you can use a single installation.
> Cheers,
> Berk
> On 10/29/12 14:47 , Carsten Kutzner wrote:
>> Dear developers,
>> we run Gromacs on an inhomogeneous cluster with different node groups -
>> old and new ones, where each group requires an individual acceleration
>> setting, e.g. SSE2 vs. SSE4.1. If I set GMX_ACCELERATION to SSE2 then I
>> won't get the optimal performance on some of the nodes, whereas with
>> SSE4.1
>> the code won't run on all nodes. One way around seems to build different
>> mdruns, and to select the optimal one after the node has been assigned by
>> the
>> queue. Is there a more elegant way like including alternate code paths,
>> where the optimal path is selected at run time?
>> 2nd question: From a performance point of view, does it hurt to have
>> support
>> for GPU, MPI, and threads compiled in - even if it is not used in many
>> cases?
>> Or would it be better to have also separate executables for that?
>> Thanks for your thoughts,
>>    Carsten
>> --
>> 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/kutzner>
> --
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/**mailman/listinfo/gmx-**developers<http://lists.gromacs.org/mailman/listinfo/gmx-developers>
> Please don't post (un)subscribe requests to the list. Use the www
> interface or send it to gmx-developers-request@**gromacs.org<gmx-developers-request at gromacs.org>
> .
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20121030/f113ff0e/attachment.html>

More information about the gromacs.org_gmx-developers mailing list