GROMACS version: 2019.2
GROMACS modification: Yes, patched with PLUMED 2.5.2
Hi everyone,
I would like to perform REST2 simulations in GROMACS. I already installed gromacs 2019.2 version patched with plumed 2.5.2. I have provided the cmake commands for the reference.
-DGMX_BUILD_OWN_FFTW=ON -DGMX_GPU=CUDA -DGMX_MPI=on -DGMX_USE_RDTSCP=ON -DGMX_SIMD=AVX2_128 -DREGRESSIONTEST_DOWNLOAD=OFF
I successfully ran one of the tutorials with five windows. I have pasted the batch file I used for this run. However, when I tried to use multiple processors per window, the simulation is getting slower. I have provided the batch file for multiple processors for comparison too.
For single processor per window:
#!/bin/bash
#SBATCH -J REST2
#SBATCH -o REST2.o%j
#SBATCH -t 72:00:00
#SBATCH -p gpu
#SBATCH -N 1 -n 5
module load gromacs/2019.2/simd_plumed
mpirun -np 5 gmx_mpi mdrun -v -deffnm rest2 -multidir rest2_* -replex 100 -plumed ./plumed.dat -hrex -cpt 600 -bonded cpu -nb cpu
For Multiple Processors:
#!/bin/bash
#SBATCH -J REST2
#SBATCH -o REST2.o%j
#SBATCH -t 72:00:00
#SBATCH -p gpu
#SBATCH -N 1 -n 10
module load gromacs/2019.2/simd_plumed
module load plumed/2.5.2
module load openmpi
mpirun -np 10 gmx_mpi mdrun -v -deffnm rest2 -multidir rest2_* -replex 100 -plumed ./plumed.dat -hrex -cpt 600 -bonded cpu -nb cpu -ntomp 25 -pin on -dlb no -pinoffset 0
I tried openMP threads 4-6 as well. However, they were slower than this option. I decided to use -ntomp 25 since I had 256 processors in the system. I also had to use -cpt 600 as my simulations was stuck at writing the cpt file and never moved past that step.
Can anyone please help me understand what I am doing wrong?
Thank you for your time,
Regards,
Nidhin Thomas