Would disabling hyperthreading improve the performance of polars?

Cloud providers such as AWS recommend disabling hyperthreading for HPC applications.

As someone who is only vaguely familiar with what that means, I am left wondering: Would disabling hyperthreading on my machine have any potential to improve the performance of polars (and, for that matter, other parallelized python libraries for data scientists)?

  • Why not just go and try it? It’s pretty easy to revert the change in the BIOS.

    – 

  • The logic is that hyper threading lets you do more work on a physical core when it’s not saturated with computational work. Since a high performance numerical library should be saturated the cores, there shouldn’t be an advantage to hyper threading. But you’d have to test to be sure.

    – 

Leave a Comment