Yeah but M2 effective speed for ML is closer to intel CPUs than it is to nvidia gpus. The correct comparison here is to compare how fast you can train on Apple M2 vs an Intel CPU with equivalent RAM.