[gmx-users] Only 2 ns/day

Kevin Boyd kevin.boyd at uconn.edu
Mon Jun 18 19:44:14 CEST 2018


One source of poor performance is certainly that you don't have SIMD
enabled. Try recompiling with SIMD enabled (the log file suggests
AVX_128_FMA). If you are compiling on gromacs on the same node
architecture that you plan to run gromacs on (and you really should be
doing this), it should detect the SIMD compatibility automatically.
See the install guide.


Also, it looks like your version of gromacs was compiled with the PGI
C and C++ compilers, which is not recommended. Consider using GCC. The
age of the compiler version matters as well - use as recent a version
as possible for performance and feature compatibility.


On Mon, Jun 18, 2018 at 6:38 PM, Alex <alexanderwien2k at gmail.com> wrote:
> Dear all,
> I use 2 nodes (each has 32 cores) for a system of 41933 atoms (containing
> water-short polymer on a solid surface). The performance is too poor, only
> 2.43 ns/day, although the imbalance is normal and around 4.9 %.
> The submission command I used is:
> gmx_mpi mdrun -ntomp 1 -deffnm eql1 -s eql1.tpr -rdd 1.5 -dds 0.9999 -npme
> 4 -ntomp_pme 1 -g eql1.log -v
> I had to use -rdd 1.5 otherwise the DD error would show up. I also tested
> manually different -npme and among all those only -npme 4 and -npme 24
> works with these 64 cores.
> I checked the -npme manually because I am not familiar with the gmx
> tune_pme.
> Below I have shared the log file of the simulation and I would be so
> appreciated if one could help me improve the performance.
> https://drive.google.com/open?id=12fX5URhvYZexST76pw3Q8wQrEPwSyzKL
> Thank you very much.
> Regards,
> Alex
> --
> 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