There is some evidence from experimentation on a test dataset that the optimize_geometries step performs best with a single core. Both the cumulative walltime (observed * n cpu) and even observed walltime are higher with more cores:
This phenomena should probably also be highlighted in the docstring
There is some evidence from experimentation on a test dataset that the optimize_geometries step performs best with a single core. Both the cumulative walltime (observed * n cpu) and even observed walltime are higher with more cores:
This phenomena should probably also be highlighted in the docstring