Just a heads up that there can be performance drop also in Stan due to patches for Meltdown bug (which need to be taken into account if someone is keeping track of speed changes between Stan versions)
TL;DR performance drop in Linux Scikit-learn & Pandas linear and logistic regression 10%, NumPy/SciPy QR decomposition 37%
That looks mostly like performance hit due to large thread numbers (Anaconda uses MKL for eg QR) where syscalls are more prevalent. Pure floating point code shouldn’t be impacted.
From what I understand (from the coverage on this week’s Security Now podcast), the cache stuff associated with meltdown should have relatively simple (and performance restoring) fixes. It’s the fixes for spectre that are going to hurt, with no clear path to anything but diminished performance for increased security.