LAMMPS is a classical molecular dynamics code, and an acronym for Large-scale Atomic/Molecular Massively Parallel Simulator. LAMMPS has potentials for solid-state materials (metals, semiconductors) and soft matter (biomolecules, polymers) and coarse-grained or mesoscopic systems.

It can be used to model atoms or, more generically, as a parallel particle simulator at the atomic, meso, or continuum scale. LAMMPS runs on single processors or in parallel using message-passing techniques and a spatial-decomposition of the simulation domain. The code is designed to be easy to modify or extend with new functionality. LAMMPS is distributed as an open source code under the terms of the GPL.

More information here.

Running LAMMPS





To see available versions when logged into Fram issue command

module spider lammps

To use LAMMPS type

module load LAMMPS/<version>

specifying one of the available versions.

Running on GPUs


We are working on compiling the LAMMPS module on Saga and Betzy with GPU support, however, since that could take some time we are, for the time being, presenting an alternative solution below.

LAMMPS is capable of running on GPUs using the Kokkos framework. However, the default distribution of LAMMPS on Saga and Betzy are not compiled with GPU support. We will therefor use singularity to use LAMMPS with GPUs.


We will first start by creating our LAMMPS singularity image based on Nvidia’s accelerated images. The following command will download the image and create a singularity container with the name lammps.sif.

[user@login-X.SAGA ~]$ singularity pull --name lammps.sif docker://

Then, to retrieve the input test file (in.lj.txt) execute the following:

[user@login-X.SAGA ~]$ wget

Slurm script

To run this we will use the following Slurm script:


#SBATCH --job-name=LAMMPS-Singularity
#SBATCH --account=nn<XXXX>k
#SBATCH --time=10:00
#SBATCH --ntasks=2
#SBATCH --mem-per-cpu=2G
#SBATCH --ntasks-per-node=4
#SBATCH --partition=accel
#SBATCH --gpus=2

set -o errexit  # Exit the script on any error
set -o nounset  # Treat any unset variables as an error

module --quiet purge  # Reset the modules to the system default
# Note: We don't need any additional modules here as Singularity is always
# available

srun singularity run --nv -B ${PWD}:/host_dir lammps.sif\
  lmp -k on g 2 -sf kk -pk kokkos cuda/aware on neigh full comm device binsize 2.8\
  -var x 8 -var y 8 -var z 8 -in /host_dir/in.lj.txt

In the script above note that --gpus=X (or --gpus-per-node=X) needs to be the same as the parameter -k on g X for LAMMPS. We also recommend that users have one MPI rank per GPU so that --ntasks=X is either equal to --gpus=X or use --gpus-per-task=1.

Performance increase

We modified the above input file to run for a bit longer (increased the number of steps to 1000). This gave us the following speed-ups compared to LAMMPS/3Mar2020-foss-2019b-Python-3.7.4-kokkos on Saga.

Node configuration

Performance (tau/day)

Speed up

40 CPU cores






2 GPUs



4 GPUs



License Information

LAMMPS is available under the GNU Public License (GPLv3). For more information, visit the LAMMPS documentation pages.

It is the user’s responsibility to make sure they adhere to the license agreements.


When publishing results obtained with the software referred to, please do check the developers web page in order to find the correct citation(s).