Hacker News Reader

That seems about on par with my experience. Even my own numeric code in C++ frequently runs significantly faster on Linux on the same hardware.

It could be something clever like maybe gcc uses a better allocator or maybe you have better, I dunno, OpenMP libraries on Linux (or whatever the C++ equivalent is)

On the other hand, when I boot into Windows for games, it always surprises me how much CPU Windows uses just… sitting there, doing nothing.

Windows has a hard time "getting out of the way" and this extends all the way down to core numerical routines. Have you ever tried installing CUDA drivers on Windows without WSL?

What blas implementation does numpy use on Windows?