b10050

ggml-org/llama.cppb10050Jul 16, 2026by github-actions[bot]same day as b10048

What changed since b10048

low risk

b10050 adds a Vulkan fix synchronizing event waits to prevent races during async transfer-queue copies.

Improvements

  • Vulkan backend now synchronizes on event_wait when transfer queues are used for asynchronous copies, tightening memory-ordering guarantees.
  • Cross-platform binary matrix (macOS, Linux, Windows, Android) remains unchanged, so existing deployment artifacts map cleanly to the new build.

Bug fixes

  • Resolved a potential race condition in Vulkan async copy handling that could cause data corruption under transfer-queue usage.

Upgrade notes

This is a small incremental build; only the Vulkan backend behavior changed, so CPU, ROCm, and OpenVINO users see no functional difference. Vulkan users running async copies should upgrade to eliminate the race, and no config or CLI changes are required.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

Fixes a potential race condition in Vulkan by synchronizing on event wait when using transfer queues for asynchronous copies.

Key Highlights

  • Fixed race condition in Vulkan async copy handling.
  • Syncs on event_wait to prevent data races during transfer queue operations.
  • Cross-platform binaries included for macOS, Linux, Windows, and Android.

Release Notes (excerpt)

<details open> vulkan: when using transfer queue for async copies, sync on event_wait to avoid race (#25229) </details> **Website:** - <https://llama.app> **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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories