b9781

ggml-org/llama.cppb9781Jun 24, 2026by github-actions[bot]same day as b9780

What changed since b9780

medium risk

b9781 adds an optional Vulkan setting to shrink graph submission batches and avoid GPU timeouts.

New features

  • Vulkan backend now exposes a knob to reduce the number of graph submission batches, mitigating timeouts on affected hardware.

Improvements

  • Build-time shader failure detection from the previous release remains in place, so broken Vulkan shaders still abort the build.

Bug fixes

  • Provides a workaround for Vulkan timeout errors that occurred when large graphs were submitted in too few batches.

Upgrade notes

No configuration migration is required; the new batch-size reduction is opt-in. Vulkan users experiencing timeouts or device-lost errors should lower the submission batch size via the new option. Default behavior is unchanged for setups that were already stable.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

Adds a configuration option to reduce the graph submission batch size in the Vulkan backend to prevent timeout issues on certain hardware configurations.

Key Highlights

  • Reduced Vulkan graph submission batches
  • Timeout prevention for Vulkan backend

Release Notes (excerpt)

<details open> vulkan: allow reducing the graph submission batches to avoid timeouts (#24872) </details> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:** - Ubuntu x64 (CPU) - Ubuntu arm64 (CPU) - Ubuntu s390x (CPU) - Ubuntu x64 (Vulkan) - Ubuntu arm64 (Vulkan) - Ubuntu x64 (ROCm 7.2) - Ubuntu x64 (OpenVINO) - Ubuntu x64 (SYCL FP32) - Ubuntu x64 (SYCL FP16) **Android:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories