Optimal Parallelisation on VSC-4

[lukas_reicht]

Dear VASP Team and VASP colleagues,

do you have experience, regarding what settings one should choose so that VASP runs optimally on the VSC-4 Cluster (https://vsc.ac.at/systems/vsc-4/)?

For the phonon calculations that I perform, large super cells are needed (in the order of 800 atoms per unit cell).

See the attached INCAR-file and job submission script for my current settings. My current parallelisation settings in the INCAR file are:
NCORE = 24
LPLANE = .TRUE.
NSIM = 4
KPAR = 1

I noticed, that most calculations do not converge when setting NCORE=48, while NCORE=24 solves that problem.

In the job submission script, I use:
#SBATCH --tasks-per-node=48

I already contacted the VSC team, which referred me to you. I am looking forward to your answers.

Best Regards,
Lukas Reicht

[ferenc_karsai]

Moved the topic here, since the VASP development group doesn't have access to the VSC-4.