The model to run on your GPU, and exactly how.

Pick your card. You get the three best models it can run, how fast each one goes, and the command that starts it. Every recipe here was run on the card and passed six checks: it loads, answers, thinks, calls tools, holds its context window and keeps pace.