[DFTB-Plus-User] choosing a good number of cores to run on

Peter Klaver klaver2 at gmail.com
Thu Dec 19 12:07:53 CET 2024


Hello people,

I’m running multiple dftb+ calculations in parallel on nodes with a pretty high number of cores and I’m wondering if parallelisation is helped by running each dftb+ calculation on a number of cores that can be factored many times.

Does it ‘help’ dftb+ to spread the work if the number of cores can be divided many times, or does it run just as efficient when the number of cores is a prime number?

And is there any difference in this between the MPI and openMP versions?

If I want to run 2 more  jobs on a node where there are 34 cores availble, should I run 2 x 17 core jobs or does it work better then to make it 2 x 16 cores?

Kind regards,
Peter 


More information about the DFTB-Plus-User mailing list