Run LAMMPS Docker Application

This guide describes how to run the LAMMPS Docker application. Go to the Application Configuration screen.

Benchmark commands

Paste the following into the Run Script text box to run the lj benchmark.

SIZE=14; STEPS=600; GPUS=8; BENCHMARK="in.lj"; cd /benchmark;
/opt/lammps/tpl/openmpi/bin/mpirun -np $GPUS /opt/lammps/bin/lmp -k on g $GPUS -sf kk -pk
kokkos cuda/aware on neigh full comm device neigh/qeq full newton on -v x $SIZE -v y $SIZE
-v z $SIZE -v steps $STEPS -in $BENCHMARK -nocite

For the eam benchmark:

SIZE=17; STEPS=300; GPUS=8; BENCHMARK="in.eam"; cd /benchmark;
/opt/lammps/tpl/openmpi/bin/mpirun -np $GPUS /opt/lammps/bin/lmp -k on g $GPUS -sf kk -pk
kokkos cuda/aware on neigh full comm device neigh/qeq full newton on -v x $SIZE -v y $SIZE
-v z $SIZE -v steps $STEPS -in $BENCHMARK -nocite

For the tersoff benchmark:

SIZE=9; STEPS=300; GPUS=8; BENCHMARK="in.tersoff";cd /benchmark;
/opt/lammps/tpl/openmpi/bin/mpirun -np $GPUS /opt/lammps/bin/lmp -k on g $GPUS -sf kk -pk
kokkos cuda/aware on neigh half binsize 5.6 comm device neigh/qeq full newton on -v x
$SIZE -v y $SIZE -v z $SIZE -v steps $STEPS -in $BENCHMARK -nocite

For the reaxff benchmark:

SIZE=48; STEPS=300; GPUS=8; BENCHMARK="in.reaxc.hns"; cd /benchmark;
/opt/lammps/tpl/openmpi/bin/mpirun -np $GPUS /opt/lammps/bin/lmp -k on g $GPUS -sf kk -pk
kokkos cuda/aware on neigh half comm device neigh/qeq full newton on -v x $SIZE -v y $SIZE
-v z $SIZE -v steps $STEPS -in $BENCHMARK -nocite

For the snap benchmark:

SIZE=23; STEPS=400; GPUS=8; BENCHMARK="in.snap.exaalt"; cd /benchmark;
/opt/lammps/tpl/openmpi/bin/mpirun -np $GPUS /opt/lammps/bin/lmp -k on g $GPUS -sf kk -pk
kokkos cuda/aware on neigh half comm device neigh/qeq full newton on -v x $SIZE -v y $SIZE
-v z $SIZE -v steps $STEPS -in $BENCHMARK -nocite

Compute the figure of merit (FOM)

From the completed Workloads list, open the completed LAMMPS job. Open the STDOUT tab. The FOM is the product of (number of atoms) × (timesteps/s).

Here’s an example stdout log from LJ run. The computed FOM is 816,175,360 atoms*timesteps/s (87808000 atoms * 9.295 timesteps/s).

Loop time of 64.5532 on 8 procs for 600 steps with 87808000 atoms
Performance: 4015.291 tau/day, 9.295 timesteps/s