Qeravio
Canonical AI event

AMD RDNA3 and RDNA4 gain int8 coopmat1 matmul implementation.

AMD RDNA3 and RDNA4 GPUs now support int8 quantized matrix multiplication in Vulkan.

25 Sept 20261 verified claims1 sources1 observations
What happened

The official source reports this update: b11160: vulkan: int8 coopmat1 matmul implementation for AMD RDNA3 and RDNA4.

Why it matters

This official update documents a development concerning b11160: vulkan: int8 coopmat1 matmul implementation for AMD RDNA3 and RDNA4. Its practical significance depends on the scope and evidence stated by the source.

What to watch next

Read the official source update and verify its stated scope, evidence and timing before acting on it.

Connected knowledge

Entities affected by this event

Llama Cppsubject
Continue this topic
infrastructureCUDA 12 and 13 now support F16 kernels for CONV_2D_DW on Ubuntu and Windows.25 Sept 2026infrastructureLlama-grammar fixes numeric truncation for token ID parsing.25 Sept 2026infrastructureHexagon multi-sequence handling added to concat_2d in llama.cpp25 Sept 2026infrastructureMetal optimizes sparse FA index caching in shared memory25 Sept 2026infrastructureHexagon adds I32 CPY and CONT support.25 Sept 2026infrastructureVulkan misalignment in conv_2d and conv_3d fixed.25 Sept 2026
Evidence trail

Sources behind the event

llama.cpp official releasesb11160: vulkan: int8 coopmat1 matmul implementation for AMD RDNA3 and RDNA4
Editorial presentation

Read the public story

Open story