Intel and AMD Collaborate to Optimize PyTorch Performance in oneDNN
Intel and AMD are working together on the oneDNN library to streamline PyTorch performance and reduce fragmentation across x86_64 Linux systems.
Intel and AMD are collaborating on software optimizations inside the oneDNN library, formerly known as MKL-DNN, to improve PyTorch workloads on Linux systems.
Engineers from both companies are addressing bottlenecks in how PyTorch interacts with low-level execution primitives. These shared upstream developments focus on reducing redundant code paths and refining how instructions are dispatched to the CPU. The joint effort cleans up implementation details that were previously fragmented between different hardware vendors.
Ultimately, these adjustments improve the execution of neural network operations across diverse x86_64 silicon. Developers using PyTorch on Linux get a consistent experience for training and inference tasks, regardless of the specific processor used, because both companies are contributing to this shared codebase.