Several versions of Gromacs are available, check below:

Compilation details:

For example Gromacs 2022.6 was patched with Plumed 2.8.3 and compiled with the GNU 8.3.0 tool chain and OpenMPI 3.1.6. The GPU enabled version was compiled with CUDA SDK (software development kit) 12.0.0 and is usable with CUDA RTL (run time library) 12.0.0. Code was generated for CUDA architecture 6.0, 7.0, 8.0.

Supported CUDA architectures:

Environment modules are available in different flavors:

To define the GMXLIB environment variable, add the following lines to the file $HOME/.bash_profile:

GMXLIB=$HOME/gromacs/top
 
export GMXLIB

Script mdrun-omp.sh to exclusively request a node and start multiple OpenMP threads (tested with –cpus-per-task=14 and –cpus-per-task=16):

mdrun-omp.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_omp
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=14
#SBATCH --exclusive
#SBATCH --time=0-24:00:00
#SBATCH --mem=120G
#SBATCH --partition=cpu
#SBATCH --qos=cpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu
module load openmpi
module load gromacs/5.1.5-cpu
 
# Set OMP_NUM_THREADS to the same value as --cpus-per-task with a fallback in   
# case it isn't set. SLURM_CPUS_PER_TASK is set to the value of --cpus-per-task,
# but only if --cpus-per-task is explicitly set.                                 
export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK:-1}
 
gmx mdrun -deffnm topology -pin on

Script mdrun-omp.sh to exclusively request a node and start multiple OpenMP threads (tested with –cpus-per-task=14 and –cpus-per-task=16):

mdrun-omp.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_omp
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=14
#SBATCH --exclusive
#SBATCH --time=0-24:00:00
#SBATCH --mem=120G
#SBATCH --partition=cpu
#SBATCH --qos=cpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu
module load openmpi
module load gromacs/2016.6-cpu
 
# Set OMP_NUM_THREADS to the same value as --cpus-per-task with a fallback in   
# case it isn't set. SLURM_CPUS_PER_TASK is set to the value of --cpus-per-task,
# but only if --cpus-per-task is explicitly set.                                 
export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK:-1}
 
gmx mdrun -deffnm topology -pin on -plumed plumed.dat

Script mdrun-mpi-omp.sh to exclusively request one or more nodes (tested with –nodes=1 and –nodes=2) and start multiple MPI processes (tested with –ntasks-per-node=6 and –ntasks-per-node=8; the number of OpenMP threads will be calculated automatically if –cpus-per-task is not explicitly set):

mdrun-mpi-omp.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_mpi_omp
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=6
#SBATCH --exclusive
#SBATCH --time=0-24:00:00
#SBATCH --mem=120G
#SBATCH --partition=cpu
#SBATCH --qos=cpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu
module load openmpi
module load gromacs/5.1.5-cpu
 
# Set OMP_NUM_THREADS to the same value as --cpus-per-task with a fallback in
# case it isn't set. SLURM_CPUS_PER_TASK is set to the value of --cpus-per-task,
# but only if --cpus-per-task is explicitly set.
if [ -n "$SLURM_CPUS_PER_TASK" ]; then
    OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK
else
    if [ -n "$SLURM_NTASKS_PER_NODE" ]; then
        OMP_NUM_THREADS=$((SLURM_CPUS_ON_NODE/SLURM_NTASKS_PER_NODE))
    else
        if [ -n "$SLURM_NTASKS" ] && [ -n "$SLURM_NNODES" ]; then
            OMP_NUM_THREADS=$((SLURM_CPUS_ON_NODE/(SLURM_NTASKS/SLURM_NNODES)))
        else
            OMP_NUM_THREADS=1
        fi
    fi
fi
 
export OMP_NUM_THREADS
 
mpirun gmx mdrun -deffnm topology -pin on

Script mdrun-mpi-omp.sh to exclusively request one or more nodes (tested with –nodes=1 and –nodes=2) and start multiple MPI processes (tested with –ntasks-per-node=6 and –ntasks-per-node=8; the number of OpenMP threads will be calculated automatically if –cpus-per-task is not explicitly set):

mdrun-mpi-omp.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_mpi_omp
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=6
#SBATCH --exclusive
#SBATCH --time=0-24:00:00
#SBATCH --mem=120G
#SBATCH --partition=cpu
#SBATCH --qos=cpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu
module load openmpi
module load gromacs/2016.6-cpu
 
# Set OMP_NUM_THREADS to the same value as --cpus-per-task with a fallback in
# case it isn't set. SLURM_CPUS_PER_TASK is set to the value of --cpus-per-task,
# but only if --cpus-per-task is explicitly set.
if [ -n "$SLURM_CPUS_PER_TASK" ]; then
    OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK
else
    if [ -n "$SLURM_NTASKS_PER_NODE" ]; then
        OMP_NUM_THREADS=$((SLURM_CPUS_ON_NODE/SLURM_NTASKS_PER_NODE))
    else
        if [ -n "$SLURM_NTASKS" ] && [ -n "$SLURM_NNODES" ]; then
            OMP_NUM_THREADS=$((SLURM_CPUS_ON_NODE/(SLURM_NTASKS/SLURM_NNODES)))
        else
            OMP_NUM_THREADS=1
        fi
    fi
fi
 
export OMP_NUM_THREADS
 
mpirun gmx mdrun -deffnm topology -pin on -plumed plumed.dat

Script mdrun-mpi-omp-9x3.sh to request one node and start 9 MPI processes and 3 OpenMP threads for each MPI process (27 threads overall):

mdrun-mpi-omp-9x3.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_mpi_omp_9x3
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=28
#SBATCH --time=0-24:00:00
#SBATCH --mem=80G
#SBATCH --partition=cpu
#SBATCH --qos=cpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu7
module load openmpi3
module load gromacs/2019.6-cpu
 
module list
 
OMP_NUM_THREADS=3
 
export OMP_NUM_THREADS
 
mpirun -np 9 gmx mdrun -deffnm meta -pin on -plumed plumed.dat

Script mdrun-mpi-omp-9x3.sh to request one node and start 9 MPI processes and 3 OpenMP threads for each MPI process (27 threads overall):

mdrun-mpi-omp-9x3.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_mpi_omp_9x3
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=28
#SBATCH --time=0-24:00:00
#SBATCH --mem=80G
#SBATCH --partition=cpu
#SBATCH --qos=cpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu8
module load openmpi3
module load gromacs/2021.7-cpu
 
module list
 
OMP_NUM_THREADS=3
 
export OMP_NUM_THREADS
 
mpirun -np 9 gmx mdrun -deffnm meta -pin on -plumed plumed.dat

The mdrun-omp-cuda.sh script runs a 4 core with 1 CUDA GPU Gromacs job:

mdrun-omp-cuda.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_omp_cuda
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
#SBATCH --gres=gpu:1
#SBATCH --time=0-24:00:00
#SBATCH --mem=16G
#SBATCH --partition=gpu
#SBATCH --qos=gpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu
module load openmpi
module load gromacs/2016.6-gpu
 
# Set OMP_NUM_THREADS to the same value as --cpus-per-task with a fallback in
# case it isn't set. SLURM_CPUS_PER_TASK is set to the value of --cpus-per-task,
# but only if --cpus-per-task is explicitly set.
export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK:-1}
 
gmx mdrun -deffnm topology -pin on -dlb auto -plumed plumed.dat

Script mdrun-mpi-omp-gpu-4x2.sh to request one node and start 4 MPI processes and 2 OpenMP threads for each MPI process (8 threads overall) with 2 CUDA GPU's:

mdrun-mpi-omp-gpu-4x2.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_mpi_omp_gpu
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=8
#SBATCH --gres=gpu:2
#SBATCH --time=0-24:00:00
#SBATCH --mem=16G
#SBATCH --partition=gpu
#SBATCH --qos=gpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu7
module load openmpi3
module load gromacs/2019.6-gpu
 
OMP_NUM_THREADS=2
 
export OMP_NUM_THREADS
 
mpirun -np 4 gmx mdrun -deffnm meta -dlb auto -plumed plumed.dat

Script mdrun-mpi-omp-gpu-4x2.sh to request one node and start 4 MPI processes and 2 OpenMP threads for each MPI process (8 threads overall) with 2 CUDA GPU's:

mdrun-mpi-omp-gpu-4x2.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_mpi_omp_gpu
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=8
#SBATCH --gres=gpu:2
#SBATCH --time=0-24:00:00
#SBATCH --mem=16G
#SBATCH --partition=gpu
#SBATCH --qos=gpu
#SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
 
module load gnu8
module load openmpi3
module load gromacs/2021.7-gpu
 
OMP_NUM_THREADS=2
 
export OMP_NUM_THREADS
 
mpirun -np 4 gmx mdrun -deffnm meta -dlb auto -plumed plumed.dat

Available versions of Gromacs:

The following examples demonstrate using the NGC GROMACS container to run the STMV benchmark.

Download the STMV benchmark:

wget https://zenodo.org/record/3893789/files/GROMACS_heterogeneous_parallelization_benchmark_info_and_systems_JCP.tar.gz 
tar xf GROMACS_heterogeneous_parallelization_benchmark_info_and_systems_JCP.tar.gz 
cd GROMACS_heterogeneous_parallelization_benchmark_info_and_systems_JCP/stmv

Script slurm-gromacs-2025.1-cpu.sh to request one node and start 8 thread-MPI tasks and 6 OpenMP threads for each thread-MPI task (48 threads overall) and run the STMV benchmark:

slurm-gromacs-2025.1-cpu.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_cpu
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=8
#SBATCH --cpus-per-task=6
#SBATCH --time=0-04:00:00
#SBATCH --mem=32G
#SBATCH --partition=cpu_guest
#SBATCH --qos=cpu_guest
##SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
test $SLURM_NNODES -eq 1 || exit 1
 
module load apptainer
module load gromacs/2025.1/cpu
 
apptainer run \
    --bind "$PWD:/host_pwd" \
    --pwd /host_pwd \
    "$GROMACS_CONTAINER" \
    gmx mdrun \
        -v \
        -ntmpi $SLURM_TASKS_PER_NODE \
        -ntomp $SLURM_CPUS_PER_TASK \
        -nb cpu \
        -pme cpu \
        -npme 1 \
        -update cpu \
        -bonded cpu \
        -nsteps 100000 \
        -resetstep 90000 \
        -noconfout \
        -dlb no \
        -nstlist 300 \
        -pin on

Script slurm-gromacs-2025.1-gpu.sh to request one node and start 8 thread-MPI tasks and 6 OpenMP threads for each thread-MPI task (48 threads thread) with 8 CUDA GPU's and run the STMV benchmark:

slurm-gromacs-2025.1-gpu.sh

#!/bin/bash --login
#SBATCH --job-name=mdrun_gpu
#SBATCH --output=%x.o%j
#SBATCH --error=%x.e%j
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=8
#SBATCH --cpus-per-task=6
#SBATCH --time=0-00:30:00
#SBATCH --mem=8G
#SBATCH --gres=gpu:a5000_mm1:8
#SBATCH --partition=gpu_guest
#SBATCH --qos=gpu_guest
##SBATCH --account=<account>
 
shopt -q login_shell || exit 1
test -n "$SLURM_NODELIST" || exit 1
test $SLURM_NNODES -eq 1 || exit 1
 
module load apptainer
module load gromacs/2025.1/gpu
 
apptainer run \
    --nv \
    --bind "$PWD:/host_pwd" \
    --pwd /host_pwd \
    "$GROMACS_CONTAINER" \
    gmx mdrun \
        -v \
        -ntmpi $SLURM_TASKS_PER_NODE \
        -ntomp $SLURM_CPUS_PER_TASK \
        -nb gpu \
        -pme gpu \
        -npme 1 \
        -update gpu \
        -bonded gpu \
        -nsteps 100000 \
        -resetstep 90000 \
        -noconfout \
        -dlb no \
        -nstlist 300 \
        -pin on