Qeravio
Canonical AI event

Gemma 4 FA tuning added for Ampere GPUs and newer.

CUDA tuning for Gemma 4 is supported on Ampere GPUs and newer. Ubuntu x64 with CUDA 13 uses libraries version 13.3.

21 Sept 20261 verified claims1 sources1 observations
What happened

The official source reports this update: b11065: CUDA: tune FA for Gemma 4 on Ampere or newer.

Why it matters

This official update documents a development concerning b11065: CUDA: tune FA for Gemma 4 on Ampere or newer. 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
infrastructureMetal kernel supports variable hc in dsv4_hc_pre20 Sept 2026infrastructureLlama.cpp now handles invalid UTF-8 sequences in its AST.20 Sept 2026infrastructureCUDA 12.8 libraries added for Ubuntu x64.20 Sept 2026infrastructureMamba time-step projection input now contiguous in llama.cpp20 Sept 2026infrastructureHexagon Adds TOP_K Operation Support20 Sept 2026infrastructureHexagon adds GEGLU_QUICK support across multiple platforms.20 Sept 2026
Evidence trail

Sources behind the event

llama.cpp official releasesb11065: CUDA: tune FA for Gemma 4 on Ampere or newer
Editorial presentation

Read the public story

Open story