Ollama Rolls Out Experimental Vulkan Support for AMD and Intel (www.phoronix.com)

Ollama has introduced experimental Vulkan GPU support in its v0.12.6-rc0 test release, enabling users who build from source to run large language models (LLMs) on GPUs via the Vulkan API. The change—tracked in a year-and-a-half long ticket and now closed with the -rc0 milestone—extends ollama’s hardware reach beyond NVIDIA/ROCm and SYCL/OpenCL pathways to include a wider range of AMD and Intel GPUs. Ollama, which leverages Llama.cpp for much of its model execution and already supports models like GPT-OSS, DeepSeek-R1, Gemma 3 and Llama 3/4, will bring Vulkan to official binaries in future releases after resolving remaining integration issues. For the AI/ML community this is significant because Vulkan provides a low-level, cross-vendor compute path that can unlock GPU acceleration on many consumer and integrated GPUs that previously had limited support. That can lower barriers to local inference, broaden hardware compatibility for developers and hobbyists, and improve deployment options on edge devices or non-NVIDIA machines. Caveats: the feature is experimental and source-build only for now, so expect rough edges and performance/compatibility tuning; community testing and feedback will be important as ollama matures Vulkan support into production binaries.