Installation Guide

System Requirements

Hardware Requirement

Verified Hardware Platforms:

  • Intel® Data Center GPU Flex Series 170

  • Intel® Data Center GPU Max Series

  • Intel® Arc™ A-Series GPUs (Experimental support)

Software Requirements

  • OS & Intel GPU Drivers

Hardware OS Driver
Intel® Data Center GPU Flex Series Ubuntu 22.04 (Validated), Red Hat 8.6 Stable 540
Intel® Data Center GPU Max Series Red Hat 8.6, Sles 15sp3/sp4 (Validated) Stable 540
Intel® Arc™ A-Series Graphics Ubuntu 22.04 Stable 540
Intel® Arc™ A-Series Graphics Windows 11 or Windows 10 21H2 (via WSL2) for Windows 11 or Windows 10 21H2
CPU (3rd and 4th Gen of Intel® Xeon® Scalable Processors) Linux* distributions with glibc>=2.17. Validated on Ubuntu 18.04. N/A

  • Intel® oneAPI Base Toolkit 2023.0

  • Python 3.7-3.10

  • Verified with GNU GCC 11

Preparations

Install Intel GPU Driver

OS Instructions for installing Intel GPU Driver
Linux* Refer to the Installation Guides for the latest driver installation for individual Linux* distributions. When installing the verified Stable 540 driver, use a specific version for component package names, such as sudo apt-get install intel-opencl-icd=22.43.24595.35
Windows 11 or Windows 10 21H2 (via WSL2) Please download drivers for Intel® Arc™ A-Series for Windows 11 or Windows 10 21H2. Please note that you would have to follow the rest of the steps in WSL2, but the drivers should be installed on Windows

Install oneAPI Base Toolkit

Please refer to Install oneAPI Base Toolkit Packages.

Need to install components of Intel® oneAPI Base Toolkit:

  • Intel® oneAPI DPC++ Compiler (DPCPP_ROOT as its installation path)

  • Intel® oneAPI Math Kernel Library (oneMKL) (MKL_ROOT as its installation path)

Default installation location {ONEAPI_ROOT} is /opt/intel/oneapi for root account, ${HOME}/intel/oneapi for other accounts. Generally, DPCPP_ROOT is {ONEAPI_ROOT}/compiler/latest, MKL_ROOT is {ONEAPI_ROOT}/mkl/latest.

NOTE: You need to activate oneAPI environment when using Intel® Extension for PyTorch* on Intel GPU.

source {ONEAPI_ROOT}/setvars.sh

NOTE: You need to activate ONLY DPC++ compiler and oneMKL environment when compiling Intel® Extension for PyTorch* from source on Intel GPU.

source {DPCPP_ROOT}/env/vars.sh
source {MKL_ROOT}/env/vars.sh

PyTorch-Intel® Extension for PyTorch* Version Mapping

Intel® Extension for PyTorch* has to work with a corresponding version of PyTorch. Here are the PyTorch versions that we support and the mapping relationship:

PyTorch Version Extension Version
v1.13.* (patches needed) v1.13.*
v1.10.* (patches needed) v1.10.*

Install via wheel files

Prebuilt wheel files availability matrix for Python versions:

Extension Version Python 3.6 Python 3.7 Python 3.8 Python 3.9 Python 3.10
1.13.10+xpu ✔️ ✔️ ✔️ ✔️
1.10.200+gpu ✔️ ✔️ ✔️ ✔️

Prebuilt wheel files for generic Python* and Intel® Distribution for Python* are released in separate repositories.

# General Python*
python -m pip install torch==1.13.0a0 torchvision==0.14.1a0 intel_extension_for_pytorch==1.13.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/

# Intel® Distribution for Python*
python -m pip install torch==1.13.0a0 torchvision==0.14.1a0 intel_extension_for_pytorch==1.13.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu-idp/us/

Note: Wheel files for Intel® Distribution for Python* only supports Python 3.9. The support starts from 1.13.10+xpu.

Note: Please install Numpy 1.22.3 under Intel® Distribution for Python*.

Note: Installation of TorchVision is optional.

Note: Since DPC++ compiler doesn’t support old C++ ABI (_GLIBCXX_USE_CXX11_ABI=0), ecosystem packages, including PyTorch and TorchVision, need to be compiled with the new C++ ABI (_GLIBCXX_USE_CXX11_ABI=1).

Note: If you need TorchAudio, please follow the instructions to compile it from source. According to torchaudio-pytorch dependency table, torchaudio 0.13.0 is recommended.

Install via compiling from source

Download source code of PyTorch and Intel® Extension for PyTorch*:

$ git clone https://github.com/pytorch/pytorch.git
$ cd pytorch
$ git checkout v1.13.0

$ git clone https://github.com/intel/intel-extension-for-pytorch.git 
$ cd intel-extension-for-pytorch
$ git checkout v1.13.10+xpu

Install PyTorch:

$ cd pytorch
$ git apply ${intel_extension_for_pytorch_directory}/torch_patches/*.patch 
$ git submodule sync
$ git submodule update --init --recursive
$ conda install numpy ninja cmake
$ pip install -r requirements.txt
$ export GLIBCXX_USE_CXX11_ABI=1
$ export USE_CUDA=0
$ python setup.py bdist_wheel
$ unset USE_CUDA
$ unset GLIBCXX_USE_CXX11_ABI
$ pip install dist/*.whl

Configure the AOT (Optional)

Please refer to AOT documentation for how to configure USE_AOT_DEVLIST. Without configuring AOT, the start-up time for processes using Intel® Extension for PyTorch* will be long, so this step is important.

Install Intel® Extension for PyTorch*:

$ cd intel-extension-for-pytorch
$ git submodule sync
$ git submodule update --init --recursive
$ pip install -r requirements.txt
$ source {DPCPP_ROOT}/env/vars.sh
$ source {MKL_ROOT}/env/vars.sh
$ export USE_AOT_DEVLIST="..." # Set values accordingly
$ python setup.py bdist_wheel
$ pip install dist/*.whl

Solutions to potential issues on WSL2

Issue Explanation
Building from source for Intel® Arc™ A-Series GPUs failed on WSL2 without any error thrown Your system probably does not have enough RAM, so Linux kernel's Out-of-memory killer got invoked. You can verify it by running dmesg on bash (WSL2 terminal). If the OOM killer had indeed killed the build process, then you can try increasing the swap-size of WSL2, and/or decreasing the number of parallel build jobs with the environment variable MAX_JOBS (by default, it's equal to the number of logical CPU cores. So, setting MAX_JOBS to 1 is a very conservative approach, which would slow things down a lot).
On WSL2, some workloads terminate with an error CL_DEVICE_NOT_FOUND after some time This is due to the TDR feature in Windows. You can try increasing TDRDelay in your Windows Registry to a large value, such as 20 (it is 2 seconds, by default), and reboot.