AIMarkTechPostabout 11 hours ago
MoonMath AI Open-Sources Faster AMD MI300X Attention Kernel
7 min read
MoonMath AI released an MIT-licensed bf16 forward attention kernel for AMD's MI300X GPU, written in HIP. The kernel beats AMD's own AITER v3 on every tested shape and rounding mode. Bare-metal access came from HotAisle, an AMD cloud provider. The kernel targets the gfx942 ISA only.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyAMD
