Setup Gromacs On AWS ParallelCluster ๐Ÿงฌ

Posted on Aug 16, 2022
tl;dr: Setup Gromacs with Spack on AWS ParallelCluster

Note: this has been turned into an AWS Workshop! ๐Ÿš€ Check it out: Gromacs on AWS ParallelCluster

Gromacs Logo

Gromacs is a popular open source Molecular Dynamics application. It supports GPU and CPU acceleration and supports multi-node processing using MPI. In the following guide we’ll setup a MPI compatible version of Gromacs using Spack package manager.


  1. In this guide, I’ll assume you already have AWS ParallelCluster Manager setup, if you don’t follow the instructions on to get started.

  2. Setup cluster with the following config gromacs-config.yaml. Some of the important options include:

Parameter Description
Shared Storage This sets up a 1.2 TB lustre drive and mounts it at /shared
HeadNode This sets up a c5a.2xlarge instance as the head node. It has 8 hyper-threaded cpus and 16 gigs of memory. This is ideal for small computational tasks such as post-processing and installing software.
ComputeNodes This sets up a queue of hpc6a.48xlarge instances. These instances have 96 physical cores and 384 GB of memory. These instances are ideal for tightly coupled compute. Note these instances don’t start running until we submit a job.
  1. Install Spack

    sudo su
    export SPACK_ROOT=/shared/spack
    mkdir -p $SPACK_ROOT
    git clone -c feature.manyFiles=true $SPACK_ROOT
    cd $SPACK_ROOT
    echo "export SPACK_ROOT=/shared/spack" >> $HOME/.bashrc
    echo "source \$SPACK_ROOT/share/spack/" >> $HOME/.bashrc
    source $HOME/.bashrc
    sudo chown -R $USER:$USER $SPACK_ROOT
  2. Setup the Spack Binary Cache to speedup the build

    spack mirror add binary_mirror
    spack buildcache keys --install --trust
  3. Next weโ€™ll use Spack to install the Intel Compilers (ICC), which we’ll use to compile gromacs.

    spack install intel-oneapi-compilers@2022.0.2

    This will take about ~4 mins to complete. Once itโ€™s complete, tell Spack about the new compiler by running:

    spack load intel-oneapi-compilers
    spack compiler find
    spack unload

    Now intel will show up as an option when we run spack compilers

    spack compilers
  4. Now that we’ve installed the intel compiler, we can proceed to install Gromacs.

    spack install -j 8 gromacs +blas +lapack %intel ^intel-oneapi-mpi
  5. This will take ~45 minutes. After it completes, we can see the installed packages with:

    spack find

    Spack Find

  6. Now we can load in gromacs and test that it works. You should see the help message from gromacs.

    spack load gromacs


  1. Download sample data sets from the Max Planck Institue in Gรถttingen

    One of the datasets we download is the benchRIB Molecule, which looks like:

    BenchRIB Dataset

    mkdir -p /shared/input/gromacs
    mkdir -p /shared/logs
    mkdir -p /shared/jobs
    cd /shared/input/gromacs
    # unzip
    unzip bench*.zip


  1. First, create a bash script gromacs.sbatch to submit jobs with:

    #SBATCH --job-name=gromacs-hpc6a-threadmpi-96x2
    #SBATCH --exclusive
    #SBATCH --output=/shared/logs/%x_%j.out
    #SBATCH --partition=hpc6a
    #SBATCH -N 2
    mkdir -p /shared/jobs/${SLURM_JOBID}
    cd /shared/jobs/${SLURM_JOBID}
    spack load gromacs
    spack load intel-oneapi-mpi@2021.5.1%intel@2021.5.0 arch=linux-amzn2-zen2
    set -x
    time mpirun -np 192 gmx_mpi mdrun -ntomp ${NTOMP} -s /shared/input/gromacs/benchRIB.tpr -resethway
  2. Submit the job:

    sbatch gromacs.sbatch
  3. We can monitor the job state with watch squeue. Once it transitions into running we’ll see:

    $ watch squeue

    From there you can ssh into one of the compute nodes and run htop to see resource consumption. If it’s running properly, you’ll see a htop output like:



  1. Install PyMol on the HeadNode

    sudo amazon-linux-extras install epel
    sudo yum update -y 
    sudo yum groupinstall -y "Development Tools"
    sudo yum install -y python3-devel glew-devel glm-devel libpng-devel libxml2-devel freetype-devel freeglut-devel qt5-qtbase
    pip3 install --user virtualenv
    virtualenv ~/hpc-ve
    source ~/hpc-ve/bin/activate
    pip3 install aws-parallelcluster==3.* nodeenv PyQt5
    nodeenv -p -n lts
    git clone
    cd pymol-open-source
    python install --prefix=~/hpc-ve/ --no-vmd-plugins
    which pymol
  2. Then open up a DCV Session from Pcluster Manager:

    Open DCV Session

  3. On the DCV session, open a terminal window and run:

    source ~/hpc-ve/bin/activate
  4. Now, in the pymol console run:

    fetch 1bl8
  5. Your output should look similar to the following:

    PyMol Output

comments powered by Disqus