My Community

When you use the same submit strategy (# of nodes, # of cores "grabbed" on each node, NPAR, and NSIM) for larger atoms that you use for smaller atoms did you see that VASP crashed? I'm starting to wonder if part of becoming good at VASP means you have to maybe make a shell script that loops throught many different node/core/NPAR/NSIM combinations for, say, a 1 ionic step run and then you see which combination of nodes, cores per node, NPAR and NSIM was the fastest. Is this part of "the game" of being good with VASP every time you "make a new POTCAR file" for a new project (for a new system). Or is there a quicker way to "psyche out" what node/core/NPAR/NSIM combination is best for any POTCAR if you know what you are doing?
Thanks in advance for the perspective!

please check the number of valence electrons and consider which combination of NPAR/NSIM is reasonable

NPAR and NSIM are for optimizing the speed. Setting them sub-optimally should not cause a run to crash. I would use a typical run to roughly tune the parameters, and assume that the results are transferable to other calculations. If you want to do a large number of similar calculations, then it worth the time to tune these parameters just for this set of calculations.

Typically, I would use all the cores in one node, unless there is not enough memory. The minimum number of nodes should correspond to the maximum memory required for your calculation. The optimal choice of nodes depends quite a lot on your system. It is a balance between the running time, efficiency, and other various reasons related to your computational resources.

If your run crashes, make sure that sufficient memory is allocated to your run. You can also work with your system administrator to see if something can be done. You can also post in the other part of the forum and see if there are specific issues with VASP for the system and compiler that you are using.