News for hackers. Stuff that matters.

A comment on Show HN: PureBee – A software-defined GPU running Llama 3.2 1B at 3.6 tok/SEC.

Re:Show HN: PureBee – A software-defined GPU running Llama 3... by smcleod on Tuesday February 24, 2026 @03:15PM (#47142293)
Managed to get 6.4tk/s with some modifications: https://github.com/PureBee/purebee/pull/1
[ Parent | on HN ]
Re:Show HN: PureBee – A software-defined GPU running Llama 3... by benryanx on Thursday February 26, 2026 @05:23PM (#47172854)
That's the postMessage bottleneck - PR #1 replaces it with Atomics-based dispatch which should push utilisation much higher. Early numbers look like 6.4 tok/sec on M2 Max
[ Parent | on HN ]
Re:Show HN: PureBee – A software-defined GPU running Llama 3... by smcleod on Friday February 27, 2026 @02:41AM (#47177726)
That's my PR
[ Parent | on HN ]