2.11.0

This release introduces the PyTorch* on Intel® Hardware container image, available for Intel CPU and XPU.

Intel® CPU support for PyTorch* provides PyTorch optimized for Intel® CPUs and takes advantage of the following instruction sets:

  • Intel® Advanced Matrix Extensions (Intel® AMX)

  • Intel® Advanced Vector Extensions 512 (Intel® AVX-512)

  • Vector Neural Network Instructions (VNNI).

Intel® GPU support is provided through the XPU device backend for supported Intel® Arc™ graphics platforms.

This image is available on Docker Hub. If you want to build it yourself, use the CPU Dockerfile or the XPU Dockerfile.

Requirements

The PyTorch* on Intel® Hardware container was validated on the host system running Ubuntu 24.04.4. It supports the following GPUs:

  • Intel® Arc™ Pro B50 Graphics

  • Intel® Arc™ Pro B60 Graphics

  • Intel® Arc™ Pro B65 Graphics

  • Intel® Arc™ Pro B70 Graphics

Components

This container is based on Intel® Open Middleware Xe (Intel® OMIX) 0.1.0 and includes PyTorch 2.11.0. For more information about this version, see the project’s Release Notes.

Known Issues

The Intel® Collective Communications Library (oneCCL) version used to build the PyTorch used in this container image (2021.17.2) does not include full optimization for PCIe collective communications. Multi-GPU use cases will not run at full performance.