[gmx-users] HPC performance in Gromacs ?
pall.szilard at gmail.com
Mon Sep 11 13:23:38 CEST 2017
More threads can help, but unless I miscounted, the number of threads you
started in your previous run was larger than the number of cores (i.e. you
oversubscribed the CPU cores). That will never be beneficial with GROMACS.
The run stats look more reasonable now, but your run still starts with
massive load imbalance:
"DD step 39 vol min/aver 1.000 load imb.: force 988.1% pme mesh/force
This is mostly likely due to the inhomogeneity in particle distribution
across the simulation space:
"There are: 17700 Atoms
Atom distribution over 54 domains: av 327 stddev 840 min 0 max 3835"
The above indicates that your system is quite inhomogeneous. Based on what
you know about your system, you can tune the decomposition to get work
distribution across ranks better. Your log file indicates that
"The maximum allowed number of cells is: X 34 Y 34 Z 34"
so the current
Domain decomposition grid 18 x 3 x 1, separate PME ranks 18
can definitely be tweaked. For instance, you can try to decompose more in y
On Wed, Sep 6, 2017 at 2:13 PM, Li, J. <jing.li at rug.nl> wrote:
> Thanks, Pall,
> I thought more threads will make the simulation faster. Is it not the case
> for Gromacs?
> I have tried to change the values. The file is provided here,
> The command line is "gmx_mpi_d mdrun -s PE_pro_50s_npt5.tpr -v -deffnm
> PE_pro_50s_npt5 -ntomp 1 -npme 18 -ntomp_pme 1 -nsteps 500 -pin on -dlb yes
> The performance is slower than before before and the imbalance is still
> high. I don't know what it means.
> Thanks in advance.
> Jing Li
> PhD student
> FMNS, University of Groningen
> Room 5115.0309, Nijenborg 4, Groningen
> The Netherlands
> 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