When Does CPU Scheduling Help AI Inference?

LLM inference is usually discussed as a GPU problem. Matrix multiplication, attention, KV-cache...