Qeravio
Canonical AI event

ggml-cuda now defaults to F32 on unsupported Nvidia and AMD devices.

ggml-cuda now supports fallback to F32 for Nvidia Ampere and AMD RDNA3 or CDNA devices.

14 Sept 20261 verified claims1 sources1 observations
What happened

The official source reports this update: b10950: ggml-cuda: fallback to F32 on device without BF16 hardware acceleration.

Why it matters

This official update documents a development concerning b10950: ggml-cuda: fallback to F32 on device without BF16 hardware acceleration. 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
infrastructureLlama.cpp moves function call to optimize decoding process14 Sept 2026infrastructureggml-cpu(s390x) repack helpers now guarded for VXE-only13 Sept 2026infrastructureLlama.cpp fixes zero divisor error in nemotron-h expert FFN size fallback.13 Sept 2026infrastructureSYCL memory error fixed in llama.cpp update13 Sept 2026infrastructureLlama.cpp applies noshuffle rule to additional row alignments.13 Sept 2026infrastructureNvidia Vulkan driver bug fixed with mutex for queuesubmit13 Sept 2026
Evidence trail

Sources behind the event

llama.cpp official releasesb10950: ggml-cuda: fallback to F32 on device without BF16 hardware acceleration
Editorial presentation

Read the public story

Open story