- Not Apple (personal choice) - As much GPU memory as possible - as low power (and long of battery life) as possible
Most of the time you just need to test your weights against a toy dataset so you don't need speed as much as a similar platform for what you really train on. And, of course, it needs to be a daily driver. That mainly means GPU mem just to have a batch size > 1 for testing. So, HN crowd, what is the best right now for low mid and high $$ point?