PyTorch 2.4 Now Supports Intel GPUs for Faster Workloads

PyTorch 2.4 has introduced support for Intel Data Center GPU Max Series, enhancing AI workloads for both training and inference. This update allows users to maintain a consistent programming experience with minimal coding changes, facilitating the integration of various hardware backends. The support includes both eager and graph modes, with optimizations for performance-critical graphs and operators through the oneAPI Deep Neural Network Library (oneDNN) and oneAPI Math Kernel Library (oneMKL). Users can easily migrate from CUDA to Intel GPUs by simply changing the device name in their code. Future updates in PyTorch 2.5 are expected to include additional Aten operators, full support for Dynamo Torchbench and TIMM benchmarks, and enhanced profiling capabilities. The community is encouraged to evaluate these new features and contribute to the ongoing development of Intel GPU support in PyTorch.

- PyTorch 2.4 now supports Intel Data Center GPU Max Series for improved AI performance.

- Users can transition from CUDA to Intel GPUs with minimal code changes.

- Future updates in PyTorch 2.5 will enhance functionality and support for additional benchmarks.

- The integration aims to provide a seamless experience across different hardware platforms.

- Community contributions are welcomed to further develop Intel GPU support in PyTorch.