[gmx-users] Number of hardware threads does not match OpenMP

[Szilárd Páll]

Hi,

The former warning is the direct consequence of the latter. Your job
scheduler or MPI launcher is setting process affinity; as a result the
OpenMP library reports 1 CPU in contrast with the 20 CPUs/cores that
our low-level detection reports. You could be loosing a considerable
amount of performance if the thread affinity you set/get is not the
correct/suitable one - my guess is that it likely is, consecutive
ranks are likely pinned to consecutive cores.

You can try either disabling the affinity setting in PBS or using some
MPI environment variable or override it with the "-pin on" mdrun
argument.

Cheers,
--
Szilárd


On Fri, Apr 10, 2015 at 8:09 PM, Dries Van Rompaey
<dries.vanrompaey at gmail.com> wrote:
> Dear all,
>
> When running a simulation of a protein in solution on my university’s cluster I get the two following notifications:
>
> Number of hardware threads detected (20) does not match the number reported by OpenMP (1).
> Consider setting the launch configuration manually!
>
> Non-default thread affinity set probably by the OpenMP library,
> disabling internal thread affinity.
>
> My PBS file is:
> #!/bin/bash
> #PBS -l nodes=2:ppn=20,pmem=2gb
> #PBS -l walltime=55:00:00
> #PBS -N GromacsmTSLP
> module load hopper/2015a
> module load GROMACS/5.0.4-intel-2015a-hybrid
>
> cd $PBS_O_WORKDIR
> mpirun gmx_mpi mdrun -s md_calcua.tpr -o $VSC_DATA/MD.trr -cpo $VSC_DATA/MD.cpt -c $VSC_DATA/MD.gro -e $VSC_DATA/MD.edr -g $VSC_DATA/MD.log
>
> The performance of the system seems to be allright (120 ns/day on 2 Xeon E5-2680 v2 @ 2.80GHz, 20 000 atoms).
> The number of processes launched seems to be correct as well (40).
> I tried looking through the mailing list, and from what I can gather these notifications should only have an effect on the system performance (which seems to be okay in my case)?
> Am I correct in thinking this won’t affect the results from my simulations?
>
> Thanks in advance
>
> Dries
> --
> 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.