4 ms·

I found ten years ago simutrans substantially improved in performance when compiled from source with -march=native and -ffast-math