[gmx-users] possible configuration for gromacs gpu compute node

Mark Abraham mark.j.abraham at gmail.com
Tue Apr 8 13:22:28 CEST 2014


On Tue, Apr 8, 2014 at 11:57 AM, Harry Mark Greenblatt <
harry.greenblatt at weizmann.ac.il> wrote:

> BS"D
>
> Dear All
>
> I was thinking of configuring some GPU compute nodes with the following:
>
> 2 x E5-2643v2 (6 cores each, 12 cores total)
> 2 x GTX 770.
>
> The idea is to run two Gromacs jobs on each node, each using 6 cores and 1
> GPU card.
>
> 1).  Does this sound reasonable.
>

Yes, assuming the existence of multiple jobs and a need to maximize overall
throughput.


> 2).  If so, I am not clear how to submit jobs so that if 1 job is running,
> the second will take the unoccupied gpu card (especially since this will be
> done under ROCKS and SGE; of course that question may be best addressed to
> the ROCKS BB).
>

mdrun -gpu_id x will specify the use of GPU x (in 0,1,...) for an mdrun
with a single domain. It is also essential that you arrange for those two
mdrun processes to have thread affinity set to cores within the same
socket, and that two mdrun processes do not compete for any core. mdrun
will respect whatever you set, but fundamentally it cannot know there's
another mdrun that it has to play nicely with. So correct use of the
scheduler and/or an intelligent job script will be needed in order to
recognise the presence of another mdrun and react correctly to it.

If the two mdrun jobs are of the same length, and you are willing to
compile with MPI and accept some runtime overhead in return for simplicity,
then mdrun_mpi -multi 2 -gpu_id 01 would manage these details
automatically, and work for two single-domain simulations occupying the
whole node.

More than one domain per simulation is possible in all cases, of course,
but you'd have to measure to see if it's useful.

Mark


> Thanks
>
> Harry
>
>
>
> -------------------------------------------------------------------------
>
> Harry M. Greenblatt
>
> Associate Staff Scientist
>
> Dept of Structural Biology           Harry.Greenblatt at weizmann.ac.il
> <mailto:arry.Greenblatt at weizmann.ac.il>
>
> Weizmann Institute of Science        Phone:  972-8-934-3625
>
> 234 Herzl St.                        Facsimile:   972-8-934-4159
>
> Rehovot, 76100
>
> Israel
>
>
>
>
>
> --
> 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.
>


More information about the gromacs.org_gmx-users mailing list