Ready-to-run recipes for Linux servers with an Intel GPU: hardware video transcoding (VAAPI / QuickSync), headless Chrome with WebGL, OpenVINO inference, OpenCL, LLM training with PyTorch XPU, Jellyfin, Immich and Frigate, on bare metal or in Docker.
They work on any Linux machine with an Intel GPU exposed as /dev/dri/renderD128:
a laptop, a mini PC, or a GPU VPS. They are tested on
Linux GPU
servers (Intel UHD 770, Debian 13).
| Example | What it does |
|---|---|
check-gpu.sh |
Checks that the GPU, VAAPI, OpenCL, Vulkan and OpenVINO all work |
ffmpeg-vaapi/ |
Hardware H.264 / HEVC transcoding with ffmpeg |
chrome-headless-gpu/ |
Headless Chrome / Playwright rendering on the GPU instead of SwiftShader |
jellyfin/ |
Jellyfin with hardware transcoding in Docker |
immich/ |
Immich photos: machine learning on the GPU (OpenVINO), Quick Sync video transcoding |
frigate/ |
Frigate NVR: VAAPI decoding + OpenVINO object detection on the GPU |
openvino/ |
Image classification on the GPU with OpenVINO, plus a benchmark |
opencl/ |
A minimal OpenCL kernel from Python |
nanochat-xpu/ |
Train Karpathy's nanochat (a small ChatGPT) on the iGPU with PyTorch XPU, optionally with the CPU |
docker/ |
Giving a container access to the GPU (Docker, Compose, Kubernetes) |
curl -fsSL https://raw.githubusercontent.com/LinuxGPU-VPS/examples/main/check-gpu.sh | bash- An Intel GPU with the
i915orxekernel driver (ls /dev/drilistsrenderD128). - Your user in the
rendergroup (sudo usermod -aG render $USER, then log in again). - On Debian / Ubuntu:
sudo apt install intel-media-va-driver-non-free vainfo ffmpeg \
intel-opencl-icd clinfo mesa-vulkan-drivers vulkan-toolsMIT. Issues and pull requests are welcome.