3 ms·
not 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
by kolinko 2y ago
not 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.