oneDNN
UXL FoundationoneDNN is a cross-platform performance library of deep-learning building blocks - convolutions, matrix multiplications, normalizations and fused primitives - implementing the oneAPI specification for the component. It is optimized for x86-64, AMD64 and 64-bit Arm CPUs and for Intel GPUs, and frameworks call it as their CPU kernel layer. The UXL Foundation hosts the project.
Openness
5 high confidence- license
- Apache-2.0(OSI)
- source
- public
- core features withheld
- no
The LICENSE file contains the Apache-2.0 text. The repository is public and the README describes no paid tier, enterprise edition or license-gated build, so the core is ungated.
- https://raw.githubusercontent.com/uxlfoundation/oneDNN/main/CMakeLists.txt recorded 2026-09-30
Top-level CMakeLists.txt builds the library from the repository and includes "cmake/options.cmake" with a "Choose the type of build" cache option. Searched the body for enterprise, license key, paid, commercial and proprietary: none.
- https://raw.githubusercontent.com/uxlfoundation/oneDNN/main/LICENSE recorded 2026-09-30
LICENSE file is the verbatim Apache-2.0 text, opening with "Apache License" and "Version 2.0, January 2004" and "TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION".
- https://raw.githubusercontent.com/uxlfoundation/oneDNN/main/README.md recorded 2026-09-30
README calls oneDNN "an open-source cross-platform" library and states "oneDNN is licensed under [Apache License Version 2.0](LICENSE)." Searched the body for enterprise edition, license key, paid tier and proprietary: none.
- https://raw.githubusercontent.com/uxlfoundation/oneDNN/main/README.md recorded 2026-09-30
README says the GPU engine "can be enabled by setting" the "ONEDNN_GPU_RUNTIME" build option and that users can "download pre-built binary packages from" conda-forge or "build the library" themselves, so every engine builds from the public source through plain options. Searched the body for enterprise edition, license key, paid tier and proprietary: none.
Adoption
2 low confidenceoneDNN is consumed as a C++ library vendored directly into frameworks rather than installed as its own package, so no download channel measures its use and GitHub stars are the only adoption signal available.
- https://api.github.com/repos/uxlfoundation/oneDNN recorded 2026-08-18
Repo metadata - stargazers_count = 4,036 - for uxlfoundation/oneDNN.
Capability
4 medium confidenceOpenVINO vendors oneDNN as two submodules, one pointing at an OpenVINO fork for the CPU plugin and one at oneapi-src/oneDNN for the GPU plugin. It supplies operators rather than the path from a framework graph to a target, which keeps it one step below Apache TVM.
- https://raw.githubusercontent.com/openvinotoolkit/openvino/master/.gitmodules recorded 2026-09-30
OpenVINO .gitmodules declares two oneDNN submodules, "src/plugins/intel_cpu/thirdparty/onednn" and "src/plugins/intel_gpu/thirdparty/onednn_gpu", the latter with "url = https://github.com/oneapi-src/oneDNN.git".
- https://raw.githubusercontent.com/uxlfoundation/oneDNN/main/README.md recorded 2026-09-30
README says "oneDNN project is part of the [UXL Foundation] and is an implementation" of the oneAPI specification, and "The library is optimized for Intel 64/AMD64 architecture based processors," "Arm(R) 64-bit Architecture (AArch64)-based processors, and Intel Graphics."
- https://raw.githubusercontent.com/uxlfoundation/oneDNN/main/README.md recorded 2026-09-30
README names the frameworks built on oneDNN, listing "ONNX Runtime", "OpenVINO(TM) toolkit" and "PyTorch\*" among them, and says oneDNN "has experimental support for the following architectures: NVIDIA\* GPU,"
Verified 2026-09-27