Skip to content
Type to start searching

    Table of Contents

    Gromacs

    Gromacs is a very efficient engine to perform molecular dynamics simulations and energy minimizations particularly for proteins. However, it can also be used to model polymers, membranes and e.g. coarse grained systems. It also comes with plenty of analysis scripts.

    Available

    • Puhti: 2018-2020 releases with regularly updated minor versions, several with plumed or cuda
    • Mahti: 2019-2020 releases with regularly updated minor versions, several with plumed
    • Check recommended version(s) with module avail gromacs-env
    • If you want to use commandline plumed tools, load the plumed module.

    Note

    We only provide the parallel version gmx_mpi, but it can be used for grompp, editconf etc. similarly to the serial version. Instead of gmx grompp ... give gmx_mpi grompp

    License

    Gromacs is free software available under LGPL, version 2.1.

    Usage

    Initialise recommended version of Gromacs on Puhti like this:

    module purge
module load gromacs-env

    Use module spider to locate other versions. To load these modules, you need to first load its dependencies, which are shown with module spider gromacs/version.

    Notes about performance

    It is important to set up the simulations properly to use resources efficiently. The most important are:

    • If you run in parallel, make a scaling test for each system - don't use more cores than is efficient. Scaling depends on many aspects of your system and used algorithms, not just size.
    • Use a recent version - there has been significant speedup over the years
    • Minimize unnecessary disk I/O - never run batch jobs with -v (the verbose flag) for mdrun
    • For large jobs, use full nodes (multiples of 40 cores, on Puhti) see example below.

    For a more complete description, consult the mdrun performance checklist on the Gromacs page.

    We recommend using the latest versions as they have most bugs fixed and tend to be faster. If you switch the major version, check that the results are comparable.

    A scaling test with a very large system (1M+ particles) may take a while to load balance optimally. It's better to increase the number of nodes in your production simulation, IF you see better performance than in the scaling test at the scaling limit, rather than run very long scaling tests in advance.

    Example parallel batch script for Puhti

    #!/bin/bash
#SBATCH --time=00:15:00
#SBATCH --partition=large
#SBATCH --ntasks-per-node=40
#SBATCH --nodes=2
#SBATCH --account=<project>
##SBATCH --mail-type=END #uncomment to get mail

# this script runs a 80 core (2 full nodes) gromacs job, requesting 15 minutes time

module purge
module load gromacs-env
export OMP_NUM_THREADS=1

srun gmx_mpi mdrun -s topol -maxh 0.2 -dlb yes

    Note

    To avoid multi node parallel jobs to spread over more nodes than necessary, don't use the --ntasks flag, but specify --nodes and --ntasks-per-node=40 to get full nodes. This minimizes communication overhead and fragmentation of node reservations. Don't use the large partition for jobs with less than 40 cores.

    Example serial batch script for Puhti

    #!/bin/bash
#SBATCH --time=00:15:00
#SBATCH --partition=small
#SBATCH --ntasks=1
#SBATCH --account=<project>
##SBATCH --mail-type=END #uncomment to get mail

# this script runs a 1 core gromacs job, requesting 15 minutes time

module purge
module load gromacs-env
export OMP_NUM_THREADS=1

srun gmx_mpi mdrun -s topol -maxh 0.2

    Example GPU script for Puhti

    #!/bin/bash
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=10
#SBATCH --gres=gpu:v100:1
#SBATCH --time=00:10:00
#SBATCH --partition=gpu
#SBATCH --account=<project>
##SBATCH --mail-type=END #uncomment to get mail

module load gromacs-env/2020-gpu

export OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK

srun gmx_mpi mdrun -s verlet -dlb yes
# additional flags, like these, may be useful - test!
# srun gmx_mpi mdrun -pin on -pme gpu -pmefft gpu -nb gpu -bonded gpu -update gpu \
    -nstlist 200 -s verlet -pin on -dlb yes

    Note

    Please make sure that using one GPU (and upto 10 cores) is at least twice as fast as using one full node of CPU cores according to the usage policy. Otherwise, don't use GPUs.

    Submit the script with sbatch script_name.sh

    Example mpi-only parallel batch script for Mahti

    #!/bin/bash
#SBATCH --time=00:15:00
#SBATCH --partition=medium
#SBATCH --ntasks-per-node=128
#SBATCH --nodes=2
#SBATCH --account=<project>
##SBATCH --mail-type=END #uncomment to get mail

# this script runs a 256 core (2 full nodes, no hyperthreading) gromacs job, requesting 15 minutes time

module purge
module load gcc/9.3.0 openmpi/4.0.3 gromacs/2020.5

export OMP_NUM_THREADS=1

srun gmx_mpi mdrun -s topol -maxh 0.2 -dlb yes

    Example mixed parallel batch script for Mahti

    #!/bin/bash
#SBATCH --time=00:15:00
#SBATCH --partition=medium
#SBATCH --ntasks-per-node=64
#SBATCH --cpus-per-task=2
#SBATCH --nodes=2
#SBATCH --account=<project>
##SBATCH --mail-type=END #uncomment to get mail

# this script runs a 256 core (2 full nodes, no hyperthreading) gromacs job, requesting 15 minutes time
# 64 tasks per node, each with 2 OpenMP threads

module purge
module load gcc/9.3.0 openmpi/4.0.3 gromacs/2020.5

export OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK

srun gmx_mpi mdrun -s topol -maxh 0.2 -dlb yes

    Visualizing trajectories and graphs

    In addition to view (not available at CSC, though) tool of Gromacs, trajectory files can be visualized with the following programs:

    • PyMOL molecular modeling system.
    • VMD visualizing program for large biomolecular systems.
    • Grace plotting graphs produced with Gromacs tools

    Note

    Please don't run visualization or heavy Gromacs tool scripts in the login node (see usage policy for details). You can run the tools in the interactive partition by prepending your gmx_mpi command with orterun -n 1, e.g. orterun -n 1 gmx_mpi msd -n index -s topol -f traj).

    References

    Cite your work with the following references:

    • GROMACS 4: Algorithms for Highly Efficient, Load-Balanced, and Scalable Molecular Simulation. Hess, B., Kutzner, C., van der Spoel, D. and Lindahl, E. J. Chem. Theory Comput., 4, 435-447 (2008).
    • GROMACS: Fast, Flexible and Free. D. van der Spoel, E. Lindahl, B. Hess, G. Groenhof, A. E. Mark and H. J. C.Berendsen, J. Comp. Chem. 26 (2005) pp. 1701-1719
    • GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, E. Lindahl SoftwareX 1 (2015) pp. 19-25
    • Tackling Exascale Software Challenges in Molecular Dynamics Simulations with GROMACS In S. Markidis & E. Laure (Eds.), Solving Software Challenges for Exascale S. Páll, M. J. Abraham, C. Kutzner, B. Hess, E. Lindahl 8759 (2015) pp. 3-27
    • GROMACS 4.5: a high-throughput and highly parallel open source molecular simulation toolkit S. Pronk, S. Páll, R. Schulz, P. Larsson, P. Bjelkmar, R. Apostolov, M. R. Shirts, J. C. Smith, P. M. Kasson, D. van der Spoel, B. Hess, and E. Lindahl Bioinformatics 29 (2013) pp. 845-54

    See your simulation log file for more detailed references for methods applied in your setup.

    More information

    Last edited Wed Jun 9 2021