Two questions before you run a local LLM: does it fit, and how fast is it really?