Getitng best performance in parallel

User discussions
1

GROMACS version: 2020.4
GROMACS modification: Yes/No
Here post your question
GROMACS version:
GROMACS modification: Yes/No
Here post your question Dear All,
I am having a cluster with 10 nodes and each nodes having 48 core processor.
I am trying to run it in 2 and 5 nodes and I am getting the maximum speed of 43 ns and 123 ns per day, which is likely slow for the above processor.
Below the script I used for running, can anyone tell me how I can achieve maximum performance here.
When I go for 10 nodes what I have to add to increase the performance?
#!/bin/sh

#PJM -j
#PJM --rsc-list “node=5”
#PJM --mpi proc=48
#PJM --rsc-list “rscunit=rscunit_ft01”
#PJM --rsc-list “elapse=12:00:00”
#PJM --rsc-list “freq=2200”
#PJM --rsc-list “group_1”

export OMP_PROC_BIND=close
export KMP_AFFINITY=verbose,compact

. /vol0001/apps/oss/spack-v0.15.4/share/spack/setup-env.sh
module load gromacs-2020.4-fj-4.2.1a-zexx6ov
export OMP_NUM_THREADS=6
mpiexec gmx_mpi mdrun -deffnm md

2

6 node and 4 MPI (export OMP_NUM_THREADS=4 )gives 70 ns (288 core)
3 node with 3 MPI gives 50 ns (144 core)
all time execution command was
mpiexec gmx_mpi mdrun -deffnm md

6 node with -ntomp 6 option I getting 93 ns (288 core)

-ntomp 12, in 12 nodes give 131 ns (576 core)
100 OpenMP thread returns following error (10000+ core)
100 OpenMP threads were requested. Since the non-bonded force buffer reduction
is prohibitively slow with more than 64 threads, we do not allow this. Use 64
or less OpenMP threads.

3

excuse me I’m stuck on this question and need to know if it can be solved or not
I need to do a simulation (REMD) for 36 replicas on one laptop that has a processor i7 with RTX 2060 (6 cores). is this suitable for simulation in parallel? or I’ll need a server or workstation if you have an answer I’ll be a pleasure and appreciate