Y
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).