[gmx-users] Running GROMACS on mini GPU cluster
szilard.pall at cbr.su.se
Mon Jul 29 20:20:43 CEST 2013
The message is perfectly normal. When you do not use all available
cores/hardware threads (seen as "CPUs" by the OS), to avoid potential
clashes, mdrun does not pin threads (i.e. it lets the OS migrate
threads). On NUMA systems (most multi-CPU machines), this will cause
performance degradation as without pinning the OS can move tightly
coupled threads belonging to the same process/MPI rank to different
NUMA regions (e.g. cores on different CPU sockets).
With four physical cores the machines in question are most probably
single-processor and for these pinning will not improve performance by
much (if at all).
On Fri, Jul 26, 2013 at 12:09 AM, Tim Moore <tcmoore3 at gmail.com> wrote:
> I am running GROMACS 4.6.2 on a GPU cluster in our reasearch group. I am trying to bench mark some systems, and I am getting this message:
> NOTE: The number of threads is not equal to the number of (logical) cores
> and the -pin option is set to auto: will not pin thread to cores.
> This can lead to significant performance degradation.
> Consider using -pin on (and -pinoffset in case you run multiple jobs).
> Each node has 2 GTX 580s and 4 physical cores. So when I submit a job I set -nt 2. But since we have a queuing system, there is no way to know which cores will be available to set -pinoffset. Does anyone have a solution, or should I even be worrying about that at all?
> gmx-users mailing list gmx-users at gromacs.org
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
More information about the gromacs.org_gmx-users