namd

Summary

Software Description

NAMD, recipient of a 2002 Gordon Bell Award, is a parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems. Based on Charm++ parallel objects, NAMD scales to hundreds of processors on high-end parallel platforms. NAMD is best run on the high performance systems where it has been compiled with the latest Intel compilers and recent MPI parallel libraries. There are different compiled versions of NAMD, which use different parallelization methods.

General Linux

OpenMPI NAMD version

The following is an example job script for running an OpenMPI-based NAMD module under SLURM:

!/bin/bash -l
#SBATCH -t 24:00:00
#SBATCH --mem-per-cpu=2gb
#SBATCH --nodes==8
#SBATCH --ntasks-per-node=32
cd $SLURM_SUBMIT_DIR

module load namd/2.14-ompi
mpirun namd2 stmv.namd > stmv.out

IBVerbs NAMD version

The following is an example of a SLURM script for running an IBVerbs version of NAMD. This script is more complex because using IBVerbs, we are responsible for telling charmrun about the topology and environment under which we want the NAMD calculation to run.

#!/bin/bash -l
#SBATCH -t 24:00:00
#SBATCH --mem-per-cpu=2gb
#SBATCH --nodes==4
#SBATCH --ntasks-per-node=8

module load namd/2.12-ibverbs
cd $SLURM_SUBMIT_DIR

# Create a formatted list of nodes
rm namd.hostfile
HOSTNAMES=$(scontrol show hostnames)
while IFS= read -r HOST;
do
 echo "host ${HOST}" >> namd.hostfile;
done <<< "$HOSTNAMES"

# Create a runscript for loading the namd module
# on each node
cat > runscript << EOL
#!/bin/bash
module load namd/2.12-ibverbs
\$*
EOL
chmod +x runscript

# Please note there is no line break
# in the following command.
`which charmrun` ++runscript ./runscript `which namd2` \
++nodelist namd.hostfile ++p $SLURM_NPROCS stmv.namd > stmv.out

One of the things that using IBVerbs allows us to do is to run NAMD on GPUs, which can offer significant performance improvements over purely-CPU-based calculations in many cases. An example job script for running NAMD on our msigpu partition follows:

#!/bin/bash -l
#SBATCH --time=2:00:00
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=24
#SBATCH --mem-per-cpu=2gb
#SBATCH -p mspigpu
#SBATCH --gres=gpu:v100:2

module load namd/2.14-libverbs-CUDA

# Create a formatted list of nodes
HOSTNAMES=$(scontrol show hostnames)
while IFS= read -r HOST;
do
 echo "host ${HOST}" >> namd.hostfile;
done <<< "$HOSTNAMES"

# Create a runscript for loading the namd module
# on each node
cat > runscript << EOL
#!/bin/bash
module load namd/2.12-ibverbs
\$*
EOL

chmod +x runscript
let NCPUS=$SLURM_CPUS_ON_NODE-1
# Please note there is no line break
# in the following command.
`which charmrun` ++runscript ./runscript `which namd2` \
++nodelist namd.hostfile ++n $SLURM_JOB_NUM_NODES ++ppn $NCPUS \
+idlepoll +devices 0,1 stmv.namd > stmv.out

Additional information may be found at: http://www.ks.uiuc.edu/Research/namd Performance hints are found at: http://www.ks.uiuc.edu/Research/namd/wiki/?NamdPerformanceTuning