[gmx-users] Regarding getting good performance for mdrun
cy16f01.dilip at nitk.edu.in
Sat May 5 08:58:41 CEST 2018
My workstation has 1 node with total 24 cores, 48 logical cores.
1] I gave the command as: gmx mdrun -v -s abc.tpr -deffnm abc
I got few messages as:-
NOTE: Parallelization is limited by the small number of atoms,
only starting 16 thread-MPI ranks.
You can use the -nt and/or -ntmpi option to optimize the number of
Using 16 MPI threads
Using 3 OpenMP threads per tMPI thread
WARNING: This run will generate roughly 51484 Mb of data
starting mdrun 'abc system'
1000000 steps, 1000.0 ps.
step 999900, remaining wall clock time: 0 s imb F 10%
Writing final coordinates.
step 1000000, remaining wall clock time: 0 s
Average load imbalance: 11.0 %
Part of the total run time spent waiting due to load imbalance: 1.7 %
NOTE: 28 % of the run time was spent communicating energies,
you might want to use the -gcom option of mdrun
2] and then if i give the command as: gmx mdrun -v -s abc.tpr -deffnm
abc -ntmpi 6 -ntomp 8
Using 6 MPI threads
Using 8 OpenMP threads per tMPI thread
step 36800, will finish Sat May 5 12:42:12 2018 imb F 6%
So, My questions are:-
a) Does the simulation run just with gmx mdrun, effect on the
b) "28 % of the run time was spent communicating...." Has it affected the
c) what does this imb F means..?? and what is the significance of the
percentage (say 1.7%) states/defines..??
c) Any other information regarding these are highly beneficial..
With Best Regards,
More information about the gromacs.org_gmx-users