llama.cpp b10215 adds Vulkan driver check to fix Intel GPU crashes

llama.cpp, the wildly popular open-source C/C++ library for running LLMs locally, just dropped a new release tagged b10215. The standout change is a Vulkan driver version check for Windows Intel GPUs, designed to eliminate the notorious crashing issues many users experienced when running models on integrated graphics. The team pinpointed the culprit as an outdated GPU driver—anything older than version 32.0.101.8860 is now flagged, and the crash guard previously used as a workaround has been removed. This is a more surgical fix: instead of gating all Intel hardware, the new logic inspects the actual driver version and only disables acceleration when necessary.

The update also refactors the GPU detection code, broadening compatibility beyond Intel's Xe2 architecture. This means Windows users with older Intel iGPUs can now safely employ Vulkan for llama.cpp inference without fear of hard crashes. The release ships with binaries for every major platform—Windows x64, Linux with CUDA 12/13, ROCm, OpenVINO, SYCL, and even Android arm64 builds. For developers and AI hobbyists running models like Llama 3 or Mistral on their own rigs, this stability patch removes a major pain point, making local inference more reliable on integrated graphics hardware.