hybrid parallelization using OPENMP+MPI

Queries about input and output files, running specific calculations, etc.


Moderators: Global Moderator, Moderator

Locked
Message
Author
musanna_galib
Newbie
Newbie
Posts: 22
Joined: Fri May 07, 2021 7:29 pm
License Nr.: 20-0330

hybrid parallelization using OPENMP+MPI

#1 Post by musanna_galib » Tue May 17, 2022 6:57 pm

Hello,

I will run some heavy Elastic band and MD simulations. I compiled VASP with MPI and OpenMP in the cluster. My question is what will be the optimum parallelization for the best performance for VASP. Do I have to mention anything in my input script? From https://www.vasp.at/wiki/index.php/Hybr ... lelization, it is not very clear to me.

For your information, I am planning for node=8:cpus=32:mpiprocs=4:ompthreads=8:mem=180.

Looking forward to hearing from you.

Thanks,
MG
MG

andreas.singraber
Global Moderator
Global Moderator
Posts: 236
Joined: Mon Apr 26, 2021 7:40 am

Re: hybrid parallelization using OPENMP+MPI

#2 Post by andreas.singraber » Fri May 20, 2022 3:25 pm

Hello!

Please understand that there is no way to tell what settings give the optimal performance without running actual benchmarks. You need to test this yourself on your machine with your settings. However, for elastic band calculations you can benchmark a single image first and then just use the same amount of cores/threads for each image. Also have a look at the parallelization information in the Wiki and our recently published video on Youtube. The hybrid parallelization overview on our Wiki you mentioned has been revised recently and is quite extensive. Please be more precise and tell us which statements are unclear. For information on how to submit hybrid jobs on your cluster, please check your cluster documentation or contact your system administrators.

All the best,

Andreas Singraber

Locked