b9995

ggml-org/llama.cppb9995Jul 14, 2026by github-actions[bot]same day as b9994

What changed since b9994

low risk

Optimized SYCL backend for Intel Battlemage architecture to improve performance.

New features

  • SYCL optimization for Intel Battlemage architecture

Improvements

  • Set vector thread count to 256 for Battlemage/Xe2 GPUs
  • Detects LunarLake, Battlemage, and XE2 architectures
  • Performance improvements on Intel Xe2 GPUs

Upgrade notes

No configuration changes are required. The binary update automatically detects the hardware architecture and applies the appropriate thread count.

Analyzed Oct 17, 2026

Upgrading?

Release notes summary

Optimized the SYCL backend for Intel's Battlemage architecture by setting the vector thread count to 256, which is preferred by these chips, while keeping the default at 128 for other devices like ARC Alchemist.

Key Highlights

  • Set `fattn_vec_nthreads` to 256 for Battlemage/Xe2 detection
  • Detects LunarLake, Battlemage, and XE2
  • Default remains 128 for ARC Alchemist
  • Improves performance on Intel Xe2 GPUs

Release Notes (excerpt)

<details open> sycl: set fattn_vec_nthreads to 256 for Battlemage (#25205) Currently detects lunarlake + battlemage / xe2 and sets the value to 256. Keeps default at 128, Intel's ARC Alchemist's prefered value. </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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories