Advice needed on optimal parallelization strategy
Posted: Wed Feb 07, 2024 1:53 pm
Dear VASP users,
I am seeking advice on the optimal parallelization strategy for running calculations with a large number of k-points and a substantial number of bands on a cluster with 192 cores per node. Despite referring to the VASP optimization guidelines, I'm still uncertain about the best approach considering my specific scenarios and seek further guidance and insights from the experienced members of this forum..
For instance, I'm attempting to perform an nscf calculation for band structures for a system with NKPTS = 800 and NBANDS = 120. I'm also doing another with NKPTS = 800 and NBANDS = 1152 (for a much larger cell).
Given these setup, I'm particularly interested in determining the appropriate values for the NCORE, KPAR, and how to set them to run the job efficiently on the 192 cores per node.
Any recommendations or insights on how to effectively distribute computational tasks across the available cores while maximizing efficiency and performance would be greatly appreciated.
Thank you in advance for your assistance
Marie
I am seeking advice on the optimal parallelization strategy for running calculations with a large number of k-points and a substantial number of bands on a cluster with 192 cores per node. Despite referring to the VASP optimization guidelines, I'm still uncertain about the best approach considering my specific scenarios and seek further guidance and insights from the experienced members of this forum..
For instance, I'm attempting to perform an nscf calculation for band structures for a system with NKPTS = 800 and NBANDS = 120. I'm also doing another with NKPTS = 800 and NBANDS = 1152 (for a much larger cell).
Given these setup, I'm particularly interested in determining the appropriate values for the NCORE, KPAR, and how to set them to run the job efficiently on the 192 cores per node.
Any recommendations or insights on how to effectively distribute computational tasks across the available cores while maximizing efficiency and performance would be greatly appreciated.
Thank you in advance for your assistance
Marie