AMD and Intel unify PyTorch CPU optimizations in oneDNN

Intel and AMD are combining their CPU optimizations for PyTorch into oneDNN, creating a unified backend for AI inference on their processors. AMD’s ZenDNN, maintained as a downstream version of oneDNN, is moving into the open-source library published by the UXL Foundation and driven by Intel. The integration is intended to let developers use a single code path while retaining hardware-specific performance tuning.

The consolidation follows a recommendation from the x86 Ecosystem Advisory Group, a joint initiative established in 2024 through which AMD and Intel collaborate. AMD has also joined the UXL Foundation as a steering member, a role that provides influence over its direction in exchange for a contribution of $20,000 per year.

oneDNN detects hardware features at runtime to generate optimized kernels and supports low-precision data types. It also accelerates common machine learning operations, including matrix multiplication and convolution. Although the collaboration centers on AMD and Intel CPUs, the library also supports Arm and RISC-V architectures, giving the shared backend a broader reach beyond x86 systems.