I looks like AMSET cannot be parallelized across more than one compute node - when I tried, I found that it was spawning all of the tasks on the first node, with the second node idling. Is this correct? If it is, updating the docs to mention it might save users some time.
Yes, unfortunately only parallelisation on a single node is supported. I will update the documentation accordingly.
Also, in case you weren’t already aware, the best performance can be obtained by setting
export OMP_NUM_THREADS=1 in your job script.