Using Intel® OMIX Container Images
Container images package software and its dependencies into a portable environment. Intel® Open Middleware Xe (Intel® OMIX) images package the Intel OMIX software stack for Intel® GPU development and runtime workloads.
This guide explains where to find Intel® OMIX container images and how to use them for development.
Available containers
Intel publishes the following containers:
Compute Runtime: The Intel® Graphics Compute Runtime for oneAPI Level Zero and OpenCL™ Driver is an open-source project that provides Level Zero and OpenCL compute API support for Intel graphics hardware.
Intel OMIX: Optimized middleware and libraries for AI workloads on Intel GPUs.
PyTorch*: Provides Intel GPU support for PyTorch on Intel® Client GPUs and Intel® Data Center GPU Max Series, integrating Intel GPUs and the SYCL software stack into the official PyTorch stack.
vLLM: A library for LLM inference and serving, optimized to accelerate inference workloads on Intel platforms.
Intel OMIX is part of a larger Intel GPU container ecosystem. The Intel Graphics Compute Runtime container provides the GPU runtime foundation. Intel OMIX builds on that foundation and adds development libraries used by GPU software and frameworks. Framework containers such as PyTorch and vLLM can then build on OMIX.
The following table provides a comparison of available variants and helps you find Docker Hub images and Dockerfiles.
Image |
Variants |
Docker Hub |
Dockerfiles |
|---|---|---|---|
Compute Runtime |
runtime, devel |
||
Intel OMIX |
runtime, devel |
||
PyTorch* |
cpu, xpu |
||
vLLM |
xpu |
Ubuntu 24.04: Link |
For a list of all supported operating systems and container image variants for each container image version, refer to the release notes.
Available image types
Available variants differ by container family:
Compute Runtime:
runtimeanddevel.Intel® OMIX:
runtimeanddevel.PyTorch:
cpuandxpu.vLLM:
xpu.
Runtime images are intended for running existing applications in production or deployment environments. They contain:
Runtime libraries
User-space GPU drivers
Compiler runtime dependencies
Components required to execute applications
Development images are intended for compiling and developing applications. They include all the components of the runtime image and additionally contain:
Intel® oneAPI DPC++/C++ Compiler
Development headers
Build tools
Library development packages
Host requirements
Intel OMIX containers require host-side GPU support. Before using a container, ensure the host has:
A supported Intel GPU and Linux distribution
A compatible kernel and Intel GPU kernel-mode driver
A container runtime configured for GPU access
Refer to the release notes for supported GPU and Linux configurations. Before deploying an OMIX container, verify that the host can detect and use the target Intel GPU.
Using images from Docker Hub
For most developers, a development image is a suitable starting point. Use the following procedure to select and run an Intel OMIX image.
Choose a development image from the
intel/omixDocker Hub tags, such asintel/omix:latest-devel-ubuntu24.04.Pull the image, replacing
<tag>in the following command with the full tag of the image.docker pull intel/omix:<tag>
Start the container with access to the host GPU:
docker run -it --rm \ --device=/dev/dri \ --ipc=host \ intel/omix:<tag>
Run
sycl-lsinside the container and check that your Intel GPU is listed. The command works in both runtime and development images.sycl-ls
Install the framework packages your application needs.
For deployment, build an image based on a runtime image, such as
intel/omix:latest-runtime-ubuntu24.04. See Docker’s multi-stage build guide for separating build and runtime stages.
Building images from source
Intel publishes Dockerfiles in the intel/containers repository. To build an image from source, clone the repository, select the appropriate Dockerfile, and run docker build from the repository root.
docker build \
-t omix-local:<version>-<variant>-<base-image> \
-f dockerfiles/omix/<version>-<variant>-<base-image>.dockerfile .
The omix-local name keeps your build separate from official intel/omix images.
Replace <version>, <variant>, and <base-image> with the matching values for the Dockerfile you want to build. For example, 0.5.0-devel-ubuntu24.04.dockerfile uses version 0.5.0, variant devel, and base image ubuntu24.04.