In our use case, from 12 cores to 2 :) (mostly for ha)