Ask HN: What's the best hardware to run small/medium models locally?

M2 MacBook? A certain graphics card? Running inference on CPU on my old Thinkpad isn't fun.