My Community

Good afternoon,

I am trying to run meta-GGAs calculations with rSCAN for a somewhat "large" system (~50 atoms).

Running band structure calculation (using the zero-weight method, outline in the hybrid functional wiki page) works well.
However, if I use noncollinear calculations with spin orbit coupling, vasp silently crashes, without any error but with a core.* file.

Given I am running it on 8 nodes, it seems to be a memory problem on the master node, since there is just a single core file.

How to deal with this problem ?
Furthermore, in general, how to reduce the memory footprint (per node) for the OpenACC version of vasp ?
The standard trick of increasing NCORE does not work, since it is always set to 1 for vasp GPU...
Furthermore, as far as I understand, KPAR does not reduce the memory footprint.

You will find attached my input files.

Thanks in advance,

Hi,

Thanks for your question. There are a few ways to reduce memory. There' a list in the wiki for saving on memory demands, and a brief overview here. A good way is to increase the number of cores.

There are a few questions for your INCAR. The way that you set MAGMOM in your INCAR file is a little unusual (https://www.vasp.at/forum/viewtopic.php?t=847). It shouldn't be an issue but it is something to clear up to make sure. For noncollinear calculations, add "LNONCOLLINEAR = T" to your INCAR file, with details given in the MAGMOM wiki page.

Best,

Chris