b8871
The latest commit solves Apple Silicon GPU hangs and adds new builds for Windows CUDA 13 and openEuler.
The open-source project llama.cpp, maintained by ggml-org, has released a significant update with commit b8871. This release is headlined by a crucial fix for macOS users, specifically a workaround for the "GPU interactivity watchdog" issue (tracked as #22216). This bug was causing system hangs on Apple Silicon (M1/M2/M3) Macs when the GPU was under sustained load during AI inference tasks. The fix ensures more stable and responsive performance when running large language models locally on Mac hardware.
The release is not just a bug fix; it's a major expansion of supported platforms. The team now provides 28 different pre-built binaries, making local AI deployment significantly easier. Key additions include new Windows builds with CUDA 13.1 DLLs for NVIDIA GPU acceleration, and specialized builds for Huawei's openEuler operating system, targeting their Ascend 310P and 910B AI accelerators with ACL Graph support. This broadens llama.cpp's reach from consumer devices to enterprise and edge computing environments where these specialized chips are deployed.
- Fixes macOS GPU watchdog bug (#22216) preventing hangs on Apple Silicon Macs during AI tasks.
- Expands to 28 pre-built binaries, adding Windows CUDA 13.1 and openEuler/Ascend accelerator support.
- Enables more stable local LLM inference across desktop, mobile (iOS/Android), and server platforms.
Why It Matters
Removes a major barrier for running LLMs on Macs and expands professional deployment options to new hardware ecosystems.