AMD and Intel have announced a collaboration to unify their software efforts for artificial intelligence on x86 processors. AMD will discontinue its separate ZenDNN library and integrate its optimizations into Intel's oneDNN project. This move aims to simplify the deployment of AI models on x86 hardware by providing a single, unified library.
The oneDNN library, which provides optimized primitives for deep learning, will now automatically select the optimal execution path for both Intel and AMD processors. This eliminates the need for developers to manage separate codebases or install specific plugins for different CPU architectures when using frameworks like PyTorch.
This initiative is the first major output from the x86 Ecosystem Advisory Group, established by the two companies in late 2024. The group also announced joint work on AI Compute Extensions (ACE), a new instruction set for matrix multiplication. Support for ACE has already been integrated into the GCC 17 compiler.
AMD has joined the governing board of the UXL Foundation, which oversees the oneDNN project under the Linux Foundation. This shift reinforces the neutral status of the library. The agreement is strictly limited to central processing units, as AMD continues to develop its independent ROCm software stack for graphics processors.
