Gromacs on the Clusters

You can run Gromacs on the cluster by using a 'container', i.e. through 'singularity'. For development or basic gromacs work, you can login to pod-gpu and work there. once you're in, type

module load singularity

to start a Gromacs container, with the option to use the GPU (gromacs can be very efficient with the GPU), type

singularity run --nv /sw/singularity/SingularityImages/gromacs-2021.sif /bin/bash

(you can 'ls /sw/singularity/SingularityImages' to see what other versions are available) this will drop you in a singularity / gromacs shell where you can do the typical gromacs commands, e.g. 'gmx'. once you're ready to run something that will take a long time (e.g. the dynamics), you should quit out of singularity (control-D) and then put the commands you need in a Slurm file, e.g.

#SBATCH -N 1 --partition=gpu --ntasks-per-node=6
#SBATCH --gres=gpu:1
##  will use the GPU partition

cd $SLURM_SUBMIT_DIR

module load singularity 

/bin/hostname
singularity run --nv  /sw/singularity/SingularityImages/gromacs-2021.sif << EOF
## Put your commands of what to do in the gromacs container  in here
#
cd run-number-1
echo "getting ready to run"
gmx mdrun  -ntmpi 1 -ntomp $SLURM_NTASKS  -nb gpu  -pin on  -s  benchRIB.tpr -nsteps 10000
date
##  End of Gromacs job commands
EOF
#  just to pause it 10 seconds...
sleep 10

and then submit and monitor with the usual 'sbatch' commands and so on.