Hacker News new | ask | show | jobs
by jtbaker 31 days ago
It's already greatly improved over previous generations due to M5s having tensor cores (higher compute capacity for matmul operations, the bottleneck for prefill).