3 ms·
they're going to be using nvidia (or maybe AMD if they ever catch up) to train these models anyways
by whimsicalism 2y ago
they're going to be using nvidia (or maybe AMD if they ever catch up) to train these models anyways
- kolinko 2y agonot necessarily so, in terms of tflops per $ (of apple’s cost of gpus, nit consumer), and tflops per watt their apple silicon is comparable if not better
- whimsicalism 2y agoflops/$ is simply not all (or even most) that matters when it comes to training LLMs.... Apple releases LLM research - all of their models are trained on nvidia.
- talldayo 2y ago> and tflops per watt their apple silicon is comparable if not better If Apple currently ships a single product with better AI performance-per-watt than Blackwell, I will eat my hat.