News for hackers. Stuff that matters.

A comment on Matrix Multiplications on GPUs Run Faster When Given "Predictable" Data. Parent comment

Re:Matrix Multiplications on GPUs Run Faster When Given "Pre... by Thorondor on Wednesday May 27, 2026 @03:21PM (#48299178)
Weights should not be 0 (at least not frequently) but in a ReLU-based neural network, activations are 0 pretty often. You're absolutely right about die area though.
[ Parent | on HN ]
Re:Matrix Multiplications on GPUs Run Faster When Given "Pre... by nomel on Wednesday May 27, 2026 @07:40PM (#48302300)
> near zero chance for you main customers workload

What percent of this hardware is running inference for ReLU models? ;)

[ Parent | on HN ]