Wiki‎ > ‎

GPU Molecular Dynamics Simulation Benchmark

posted May 17, 2011, 6:12 PM by Dong Xu   [ updated Sep 13, 2013, 4:52 PM ]
ACEMD

acemd input

NAMD

Multi-core CPU-only, 4-core job
namd2_mc +p4 dhfr.namd

GPU+CPU job  (Max 14 CPU cores can be used to share 1 GPU on chili)
charmrun_gpu ++local +p4 /home/apps/bin/namd2_gpu +idlepoll dhfr.namd
nohup namd2_cuda +idlepoll +p8 +devices 0,2 mineq-01.conf >&mineq-01.conf.log


To loop thru different number of CPU cores:

for i in {2..4}; do echo $i; charmrun_gpu ++local +p$i /home/apps/bin/namd2_gpu +idlepoll dhfr.namd > p${i}g1.log; done


nohup sh -c 'ls -d */|cut -d / -f1|xargs -ti sh -c "cd {}; echo {}; run_namd_gpu.sh 1 48 1 {}.namd; cd .."' >& chili.log &  (Using the run_namd_gpu.sh script!)

grep day *.log
and take the day/ns value from the last output: Info: Benchmark time: .......

More info on NAMDWIKI

http://www.ks.uiuc.edu/Research/namd/wiki/index.cgi?Namd28b2Release

http://www.ks.uiuc.edu/Research/namd/2.8/notes.html

GROMACS

mdrun_gpu -s topol.tpr (using gromacs dhfr test suite)

NAMD support multi-CPU and multi-GPU, ACEMD (free version) and GROMACS only support 1CPU+1GPU

Some benchmarks:

http://biowulf.nih.gov/apps/gromacs-gpu.html
Comments