PyTorch Prerequisites for Intel® GPUs

Overview

This guide provides instructions for installing the prerequisites needed to run and build PyTorch 2.15 on Intel GPUs.

If you are compiling and using PyTorch 2.14, please refer to the prerequisite instructions specific to PyTorch 2.14.

Choose your installation path

The prerequisites are organized by operating systems. Pick the section that matches your system:

Linux — Install Intel® Open Middleware Xe (Intel® OMIX). A single package set installs the complete GPU software stack needed to run and build PyTorch, so no separate driver or Intel® Deep Learning Essentials installation is required. If your GPU or OS is not yet covered by Intel® OMIX (including Intel® Data Center GPU Max Series), install the Intel GPU driver and Intel® Deep Learning Essentials separately.
Microsoft* Windows — Install the Intel GPU driver and Intel® Deep Learning Essentials separately.
Microsoft* Windows Subsystem for Linux 2 (WSL 2) — The Windows host driver owns the GPU, so follow the Windows driver steps on the host and install the Intel user-mode compute runtime (and Intel® Deep Learning Essentials, only if you need to build from source) in the WSL2 Linux guest environment.

Linux

Install Intel® Open Middleware Xe (Intel® OMIX)

Intel® OMIX is a validated Intel GPU software stack with pinned component versions. One package set installs the complete middleware stack needed for PyTorch binary use or build from source:

Intel Compute Runtime
Intel performance libraries:
- Intel® oneAPI DPC++ Library (oneDPL)
- Intel® oneAPI Math Kernel Library (oneMKL)
- Intel® Collective Communications Library (oneCCL)
- Intel® Profiling Tools Interfaces for GPU

PyTorch 2.15 is validated against Intel® OMIX 0.4.0. If you need to build an earlier PyTorch release from source, use the matching Intel® OMIX version.

PyTorch version	Intel® OMIX version
2.15	0.4.0
2.14	0.3.0

To install Intel® OMIX, follow the steps at Installing Intel® Open Middleware Xe.

Set Up the OMIX Environment Variables

Note If you installed PyTorch from binaries (pip wheels), the required runtime libraries already come from the wheel dependencies. Do not source the script below, to avoid version conflicts.

For building PyTorch from source and for running the wheel you built, source the oneAPI environment script provided by intel-omix-dev:

source /opt/intel/oneapi/setvars.sh

Consider adding this to your ~/.bashrc so it runs every time you log in or create a new shell session.

Then verify the installation:

xpu-smi discovery
clinfo -l
source /opt/intel/oneapi/setvars.sh && sycl-ls

Intel® Data Center GPU (Max Series)

Intel® Data Center GPU Max Series is not yet covered by Intel® OMIX. Install the Intel GPU driver and Intel® Deep Learning Essentials separately.

Step 1: Install the Intel GPU driver

The Data Center GPU Driver LTS2 Installation Instructions describe software installation for Intel® Data Center GPU Max Series systems, along with compute and media runtimes and development packages.

Optionally, follow these instructions to verify expected Intel GPU hardware is working.

Step 2: Install Intel® Deep Learning Essentials

Skip this step if you only run PyTorch wheel binaries. It is required only to build PyTorch from source.

Option A: Install using YUM/Zypper package manager

For RPM-based distributions such as Red Hat Enterprise Linux Server, YUM is the usual choice. You'll need to configure YUM to install software packages that aren't available in the default repositories. These instructions show how to add access to the appropriate Intel repository, along with the public key used to authenticate the downloaded packages.

Create an Intel YUM repository information file and move it to the YUM configuration directory:

tee > /tmp/oneAPI.repo << EOF
[oneAPI]
name=Intel® oneAPI repository
baseurl=https://yum.repos.intel.com/oneapi
enabled=1
gpgcheck=1
repo_gpgcheck=1
gpgkey=https://yum.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
EOF
sudo mv /tmp/oneAPI.repo /etc/yum.repos.d

Use YUM/Zypper to install Intel® Deep Learning Essentials:

sudo yum install intel-deep-learning-essentials-2026.1
or
sudo zypper install intel-deep-learning-essentials-2026.1

Option B: Install using APT package manager (Ubuntu)

For Debian-based Linux distributions such as Ubuntu Server, APT is the usual choice. You'll need to configure APT to install software packages that aren't available in the default repositories. These instructions show how to add access to the appropriate Intel repository, along with the public key used to authenticate the downloaded packages.

Make sure the necessary tools to add repository access are available:
```
sudo apt update
sudo apt install -y gpg-agent wget gnupg
```

Download the Intel APT repository's public key and put it into the /usr/share/keyrings directory:

wget -qO- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | sudo gpg --dearmor -o /usr/share/keyrings/oneapi-archive-keyring.gpg
echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list

Update the APT package list and repository index:
```
sudo apt update
```

Use APT to install Intel® Deep Learning Essentials:

sudo apt install -y intel-deep-learning-essentials-2026.1

Option C: Install using offline installation scripts

Instead of using a package manager, you can install Intel® Deep Learning Essentials using offline installation scripts. Each installation script is a file containing all the needed files together with a script that extracts and installs the development package.

Important Use sudo to install files in system directories so they're available globally. Without sudo, files are installed in the current user's home directory.

Make sure the necessary tools are available:
```
sudo apt update
sudo apt install -y wget
```

Download the Intel® Deep Learning Essentials offline installation script and install:

wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/c109e1ae-e02c-48a6-917b-b03b90d33f77/intel-deep-learning-essentials-2026.1.2.25_offline.sh
sudo sh ./intel-deep-learning-essentials-2026.1.2.25_offline.sh -a --silent --eula accept

Set Up Intel Deep Learning Environment Variables

For building PyTorch from source, use the following commands to configure the build environment:

source /opt/intel/oneapi/compiler/latest/env/vars.sh
source /opt/intel/oneapi/umf/latest/env/vars.sh
source /opt/intel/oneapi/pti/latest/env/vars.sh
source /opt/intel/oneapi/ccl/latest/env/vars.sh
source /opt/intel/oneapi/mpi/latest/env/vars.sh

For running PyTorch after the wheel is built, source the above variables and additionally source tcm:

source /opt/intel/oneapi/compiler/latest/env/vars.sh
source /opt/intel/oneapi/umf/latest/env/vars.sh
source /opt/intel/oneapi/pti/latest/env/vars.sh
source /opt/intel/oneapi/ccl/latest/env/vars.sh
source /opt/intel/oneapi/mpi/latest/env/vars.sh
source /opt/intel/oneapi/tcm/latest/env/vars.sh

Consider adding the runtime commands to your ~/.bashrc file so they run every time you log in or create a new shell session.

Windows

On Windows, install the Intel GPU driver and Intel® Deep Learning Essentials separately.

Step 1: Install the Intel GPU driver for Windows

Follow the instructions in the Intel® & Iris® Xe Graphics - Windows documentation to download and run the installer to update your WHQL Certified graphics driver to version 32.0.101.8801(Latest) or higher. Please include LevelZeroSDK in the installation package for torch.compile usage on Windows.

Hardware verified with Windows 11

Intel® Arc™ A-Series Graphics
Intel® Arc™ B-Series Graphics
Intel® Core™ Ultra Processors with Intel® Arc™ Graphics
Intel® Core™ Ultra Mobile Processors (Series 2) with Intel® Arc™ Graphics
Intel® Core™ Ultra Processors (Series 2) with Intel® Arc™ Graphics
Intel® Core™ Ultra Mobile Processors (Series 3) with Intel® Arc™ Graphics

If you only need to run PyTorch on Intel GPU using the official PyTorch wheels, follow Get Started with PyTorch on Intel GPUs — no further steps on this page are needed. If you need to build PyTorch from source on Windows, continue to Step 2 below.

Step 2: Install Intel® Deep Learning Essentials for Windows

Skip this step if you only run PyTorch wheel binaries. It is required only to build PyTorch from source. PyTorch 2.15 requires version 2026.1.3.

Click on the following to download the Intel® Deep Learning Essentials package. Then double-click on the downloaded exe file to run it and follow the instructions to install: intel-deep-learning-essentials-2026.1.3.20_offline.exe.

Set Up Intel Deep Learning Environment Variables

Use this command to configure environment variables, important folders, and command settings.

call "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\env\vars.bat"
call "C:\Program Files (x86)\Intel\oneAPI\ocloc\latest\env\vars.bat"

These commands must be run every time you log in or create a new shell session.

WSL2

Windows Subsystem for Linux 2 runs a Linux distribution as a guest on a Windows host. The driver stack is split: the Windows host driver owns the GPU and exposes it to the guest, and the guest only needs the Intel user-mode compute runtime (UMD). No Linux kernel-mode GPU driver is installed inside the WSL2 guest.

Hardware verified with WSL2 (Ubuntu 26.04 & 24.04)

Intel® Arc™ A-Series Graphics
Intel® Arc™ B-Series Graphics
Intel® Core™ Ultra Processors with Intel® Arc™ Graphics
Intel® Core™ Ultra Mobile Processors (Series 2) with Intel® Arc™ Graphics
Intel® Core™ Ultra Processors (Series 2) with Intel® Arc™ Graphics
Intel® Core™ Ultra Mobile Processors (Series 3) with Intel® Arc™ Graphics

Step 1: Set up the host and guest driver stack

On Windows host: Install the Windows GPU driver, as described in Install the Intel GPU driver for Windows.
On Windows host: List the Ubuntu distributions available for installation, then install a distribution from the validated list:
```
wsl --list --online
wsl --install -d Ubuntu-26.04
```
On WSL2 guest: Add the Intel client GPU package repository, following the Client GPU installation instructions for latest Ubuntu. Follow only the repository setup steps.

On WSL2 guest: Install the user-mode compute runtime packages:

sudo apt update
sudo apt install -y libze1 libze-intel-gpu1 intel-opencl-icd clinfo libze-dev intel-ocloc

On WSL2 guest: Verify that the guest sees the GPU:
```
ls -l /dev/dxg
clinfo -l
```
The /dev/dxg must exist, and clinfo -l must list an Intel platform with at least one device.

If /dev/dxg is missing, the host is not exposing the GPU. Update the Windows host driver and restart WSL2:
```
wsl --shutdown
wsl -d Ubuntu-26.04
```
If /dev/dxg exists but no device is listed, the runtime packages from step 4 are missing or incomplete.

Step 2: Install Intel® Deep Learning Essentials in the guest

Skip this step if you only run PyTorch wheel binaries. It is required only to build PyTorch from source.

Make sure the necessary tools to add repository access are available:
```
sudo apt update
sudo apt install -y gpg-agent wget gnupg
```

Download the Intel APT repository's public key and put it into the /usr/share/keyrings directory:

wget -qO- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | sudo gpg --dearmor -o /usr/share/keyrings/oneapi-archive-keyring.gpg
echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list

Update the APT package list and repository index:
```
sudo apt update
```

Use APT to install Intel® Deep Learning Essentials 2026.1:

sudo apt install -y intel-deep-learning-essentials-2026.1

Set Up Intel Deep Learning Environment Variables

Note If you installed PyTorch from binaries (pip wheels), the required runtime libraries already come from the wheel dependencies. Do not source the variables below, to avoid version conflicts.

For building PyTorch from source, use the following commands to configure the build environment:

source /opt/intel/oneapi/compiler/latest/env/vars.sh
source /opt/intel/oneapi/pti/latest/env/vars.sh
source /opt/intel/oneapi/umf/latest/env/vars.sh
source /opt/intel/oneapi/ccl/latest/env/vars.sh
source /opt/intel/oneapi/mpi/latest/env/vars.sh

For running the wheel you built, source the above variables and additionally source tcm:

source /opt/intel/oneapi/compiler/latest/env/vars.sh
source /opt/intel/oneapi/pti/latest/env/vars.sh
source /opt/intel/oneapi/umf/latest/env/vars.sh
source /opt/intel/oneapi/ccl/latest/env/vars.sh
source /opt/intel/oneapi/mpi/latest/env/vars.sh
source /opt/intel/oneapi/tcm/latest/env/vars.sh

Consider adding the runtime commands to your ~/.bashrc so they run every time you log in or create a new shell session.

Where to go next?

After installing the prerequisites as shown above, you're ready to return to and continue following the upstream PyTorch instructions in the PyTorch Building from Source: Install Dependencies section.

Select Your Language

Using Intel.com Search

Quick Links

Recent Searches

Advanced Search

Only search in

PyTorch* Prerequisites for Intel® GPUs

Overview

Choose your installation path

Linux

Install Intel® Open Middleware Xe (Intel® OMIX)

Set Up the OMIX Environment Variables

Intel® Data Center GPU (Max Series)

Step 1: Install the Intel GPU driver

Step 2: Install Intel® Deep Learning Essentials

Option A: Install using YUM/Zypper package manager

Option B: Install using APT package manager (Ubuntu)

Option C: Install using offline installation scripts

Set Up Intel Deep Learning Environment Variables

Windows

Step 1: Install the Intel GPU driver for Windows

Step 2: Install Intel® Deep Learning Essentials for Windows

Set Up Intel Deep Learning Environment Variables

WSL2

Step 1: Set up the host and guest driver stack

Step 2: Install Intel® Deep Learning Essentials in the guest

Set Up Intel Deep Learning Environment Variables

Where to go next?

Product and Performance Information