Scaling/performance of autodE about autode HOT 2 CLOSED

This is an excellent question and one that I've not really explored at all. My default was to use 8 cores for small things or 16 for anything with >20 atoms or so, mostly due to the scaling of QM codes generally not being great past 16 cores. With ORCA v.4 at least there's not close to 2x speedup going 8→16 cores (and only gets worse from there). The rational being that finding TSs is generally serial in the calculations that are performed (with the exception of NEB, which isn't the default approach). I suppose it rather depends on how much compute you have access to!

I think your suggested formula is a pretty good start, definitely with an upper limit! However, there was an interesting paper from Hutchinson which suggests the number of conformers grows as log rather than exponentially

from autode.

Oh, ok thank you for the paper. If we quickly discard higher energy ones, it seems to be subexponentially, alright. I see, so with the TS search probably being the limiting factor it comes back down to the usual single calculation requirements.

from autode.