Default install only uses a single thread

[Jimmy-Z]

Probably nobody wants this.

I first tried the default install without pytorch/cuda, thinking maybe it's fast enough, turns out it's not, and I noticed it's single thread. then I installed pytorch and tried cuda, sure it's fast, but only like 10x, so I tried --force-cpu, thinking maybe installing pytorch will enable multi-threading, turns out it did, and performance is comparable to cuda mode, perf/power wise it's better.