TherMOF

Running Lammps

How to make your Lammps run faster

I did some benchmarking to see how I can improve Lammps simulation speed. The sources above were quite helpful. I tried using srun as well as mpirun and I also tried using the export OMP_NUM_THREADS command. For all trials I used mpi cluster in H2P with 2 nodes and 8 processors per node. For me the fastest was just using mpirun with total number of tasks:

mpirun -np ${SLURM_NTASKS} lmp_mpi -in in.MOF5 > lammps_out.txt

>>> Total wall time: 0:08:08

Close second was assigning the OMP_NUM_THREADS variable:

export OMP_NUM_THREADS=2
mpirun -np 8 lmp_mpi -in in.MOF5 > lammps_out.txt

>>> Total wall time: 0:09:49

The tests were done for calculating thermal conductivity of MOF5. For other types of simulations results may differ.