[gmx-users] Recommendations on how to increase performance
Yunlong Liu
yliu120 at jhmi.edu
Mon Aug 18 19:18:55 CEST 2014
Hi Gromacs Users,
I got an time accounting table from a single run. Can anyone give me any advice on how to further increase performance?
R E A L C Y C L E A N D T I M E A C C O U N T I N G
On 16 MPI ranks, each using 8 OpenMP threads
Computing: Num Num Call Wall time Giga-Cycles
Nodes Threads Count (s) total sum %
-----------------------------------------------------------------------------
Domain decomp. 16 8 10001 42.603 14723.278 3.3
DD comm. load 16 8 2001 0.034 11.807 0.0
Neighbor search 16 8 10001 50.194 17346.533 3.9
Comm. coord. 16 8 190000 33.143 11454.105 2.6
Force 16 8 200001 530.787 183435.628 41.7
Wait + Comm. F 16 8 200001 30.357 10491.253 2.4
PME mesh 16 8 200001 491.803 169963.077 38.7
NB X/F buffer ops. 16 8 580001 12.216 4221.874 1.0
Write traj. 16 8 4 0.081 28.145 0.0
Update 16 8 200001 5.612 1939.518 0.4
Constraints 16 8 200001 67.633 23373.282 5.3
Comm. energies 16 8 20001 2.636 910.895 0.2
Rest 4.649 1606.573 0.4
-----------------------------------------------------------------------------
Total 1271.748 439505.968 100.0
-----------------------------------------------------------------------------
Breakdown of PME mesh computation
-----------------------------------------------------------------------------
PME redist. X/F 16 8 400002 135.738 46909.937 10.7
PME spread/gather 16 8 400002 119.991 41467.827 9.4
PME 3D-FFT 16 8 400002 59.470 20552.340 4.7
PME 3D-FFT Comm. 16 8 800004 170.363 58876.121 13.4
PME solve Elec 16 8 200001 5.465 1888.524 0.4
-----------------------------------------------------------------------------
Thank you.
Yunlong
Davis Yunlong Liu
BCMB - First Year PhD Candidate
School of Medicine
The Johns Hopkins University
E-mail: yliu120 at jhmi.edu<mailto:yliu120 at jhmi.edu>
More information about the gromacs.org_gmx-users
mailing list