Eric Quinnell(@divBy_zero):mxfp8 x mxfp4 at higher perf than normal mxfp8 matmul is a magic bullet worth pointing out. Kudos to Richard Ho and the team, this is good engineering. That’s 2x activation fetches plus scale sidecar concurrently on the fp8 to match the fp4 native element sizes. Not trivial

Eric Quinnell

@divBy_zero

Founding Chip Architect, @river_ai_inc Fmr AWS Tranium, Tesla Dojo, CPUs (x86+ARM). PhD Computer Arithmetic

参加 December 2022

471 フォロー中 2.7K ファン

Eric Quinnell@divBy_zero

2026.08.27 13:07

mxfp8 x mxfp4 at higher perf than normal mxfp8 matmul is a magic bullet worth pointing out. Kudos to Richard Ho and the team, this is good engineering. That’s 2x activation fetches plus scale sidecar concurrently on the fp8 to match the fp4 native element sizes. Not trivial