Submission scripts for our local supercomputers
We are using both the Arina and Atlas clusters for running jobs using the Gromacs software package. Below we show PBS scripts for both supercomputers.
KALK
#!/bin/bash
#PBS -l nodes=nd15:ppn=16
#PBS -l walltime=480:00:00
#PBS -joe
#PBS -q gpu2
## Do not change below this line
. /home/users/etc/send_funcs.sh
HOST=$(hostname)
echo Job runing on $HOST
echo $PBS_O_WORKDIR > /home/qsub_priv/pbs_paths/PBS.$PBS_JOBID.$USER
chmod 755 /home/qsub_priv/pbs_paths/PBS.$PBS_JOBID.$USER
cd $PBS_O_WORKDIR
chooseScr
cd $scr
#echo "cp -r $PBS_O_WORKDIR/* $scr/."
cp -r $PBS_O_WORKDIR/*tpr $scr/.
#source /software/gromacs/bin/GMXRC.bash
#module load GROMACS/2019.4-fosscuda-2018b
module load GROMACS/2020-fosscuda-2018b
tpr="topol.tpr"
/usr/bin/time -p mpirun -np 1 gmx_mpi mdrun -ntomp 16 -quiet -nb gpu -gpu_id 01 -s $tpr > $PBS_JOBID.out
recover
rm -f /home/qsub_priv/pbs_paths/PBS.$PBS_JOBID.$USER
ATLAS-FDR
#!/bin/bash
#PBS -q qchem
#PBS -l nodes=1:ppn=16
#PBS -l mem=1gb
#PBS -l cput=1000:00:00
#PBS -N qchem
module load GROMACS/2019.4-foss-2019b
cd $PBS_O_WORKDIR
export NPROCS=`wc -l < $PBS_NODEFILE`
tpr="topol.tpr"
np=8
ntomp=2
gmx mdrun -ntmpi $np -ntomp $ntomp -s $tpr
ATLAS-EDR
We are also using the extension of the Atlas supercomputer, which uses SLURM instead of PBS. A submission script for Gromacs looks as follows. You can check the documentation for options in Slurm here.
#!/bin/bash
#SBATCH --partition=long
#SBATCH --job-name=hase
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=4
#SBATCH --cpus-per-task=1
#SBATCH --gres=gpu:p40:1
#SBATCH --time=1-00:00:00
#SBATCH --mem=64G
module load GROMACS/2020-fosscuda-2019b
tpr="topol.tpr"
np=8
ntomp=1
ngpu=2
mpirun --map-by ppr:4:node -np $np gmx_mpi mdrun -ntomp $ntomp -s $tpr