[gmx-users] GPU performance

[RAHUL SURESH]

Hi users

I generated my 200ns production Tpr file in 2016 version and running it on
version 2018, in the cluster supported by gpu.
32 core and 1 P100 GPU card.

In the output file, I get a message stating " Multiple energy groups is not
implemented for GPUs, falling back to the CPU." What is this about and how
to overcome it.?

Also, the output files are not being generated simultaneously in the
working directory.

*script as follows*

*#!/bin/bash*
*#PBS -S /bin/bash*
*#PBS -N gromax-alz*
*#PBS -l select=1:ncpus=32:mpiprocs=1:ompthreads=32:ngpus=1*
*#PBS -l walltime=360:00:00*
*#PBS -q gpu*
*#PBS -joe*
*#PBS -V*

*#export I_MPI_FABRICS=shm:tmi*
*#export I_MPI_PROVIDER=psm2*
*#export I_MPI_FALLBACK=0*

*cd $PBS_O_WORKDIR*
*cat $PBS_NODEFILE > pbs_nodes*
*echo Working directory is $PBS_O_WORKDIR*
*NPROCS=`wc -l < $PBS_NODEFILE`*
*NNODES=`uniq $PBS_NODEFILE | wc -l`*


*if [ "$NNODES" == "1" ]*
*then*
*jid=`echo $PBS_JOBID | awk -F'.' '{print $1}'`*
*mkdir /scratch/$jid/*
*export TMPDIR=/scratch/$jid/*
*else*
*jid=`echo $PBS_JOBID | awk -F'.' '{print $1}'`*
*usern=`echo $USER`*
*mkdir -p /home/$usern/scratch/$jid*
*export TMPDIR=/home/$usern/scratch/$jid*
*fi*

*cd $TMPDIR*

*export IntelPath=/apps/intel2017u7*
*source $IntelPath/bin/compilervars.sh intel64*
*source $IntelPath/mkl/bin/mklvars.sh intel64*
*source $IntelPath/impi/*/bin64/mpivars.sh intel64*
*source /apps/batch-scripts/cuda92.sh*

*source /apps/gromacs/gpu/2018.2/intel/2017u7/cuda/9.2/bin/GMXRC*

*cp $PBS_O_WORKDIR/alz.tpr $TMPDIR/*


*gmx_mpi mdrun -v -deffnm alz 2>&1 | tee $PBS_O_WORKDIR/OUTPUT.txt*


*mkdir $PBS_O_WORKDIR/out*

*cp --force $TMPDIR/* $PBS_O_WORKDIR/out/*

*rm -rf $TMPDIR*
*date*


-- 
*Regards,*
*Rahul *