# uxlfoundation/oneDNN

oneAPI Deep Neural Network Library (oneDNN)

Repository: https://github.com/uxlfoundation/oneDNN
Canonical: https://ross.abutalabs.com/products/onednn
Homepage: http://uxlfoundation.github.io/oneDNN/
Language: C++
License: Apache-2.0
License Family: permissive
Topics: onednn, oneapi, deep-learning, deep-neural-networks, performance, cpp, openmp, tbb, x86-64, x64, aarch64, avx512, amx, xe-architecture, library, bfloat16, sycl, vnni
Last push: 2026-08-26T22:54:50+00:00

## Health v2 (maintenance only)
Score: 99/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 99, longevity 100
- inputs: {"age_days": 3768, "days_push": 7, "days_rel": 7, "gap_med": 18, "n_releases_24m": 30}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4042, forks 1186 (observed 2026-08-28T04:08:33.530932+00:00)

## What it is
oneDNN is an open-source cross-platform performance library providing optimized building blocks (primitives) for deep learning applications on CPUs and GPUs. It implements the oneAPI specification and is optimized for Intel 64/AMD64 processors, Arm AArch64 processors, and Intel Graphics, with experimental support for NVIDIA/AMD GPUs, PPC64, s390x, and RISC-V.

## Use cases
- speed up convolution and matmul primitives in my deep learning framework
- optimize inference performance on Intel CPUs and GPUs
- integrate accelerated neural network kernels into a C++ application
- leverage AVX-512, AMX, and bfloat16 hardware features for DNN workloads
- accelerate PyTorch or TensorFlow execution on Intel architecture
- build a custom inference engine with optimized low-level primitives

## When to choose
- you are developing a deep learning framework or inference engine and need highly optimized CPU/GPU primitives
- you target Intel hardware (including AMX, AVX-512, Intel Graphics) or Arm AArch64 and want vendor-tuned kernels
- you need a oneAPI-spec-compliant DNN library with SYCL support
- you maintain software like llama.cpp, ONNX Runtime, or OpenVINO-style stacks needing a fast backend

## When to avoid
- you are a practitioner who just wants to train or run models - use PyTorch, TensorFlow, or another oneDNN-enabled framework instead
- you need high-level model APIs, automatic differentiation, or training loops - oneDNN provides low-level primitives only
- your primary targets are NVIDIA/AMD GPUs or RISC-V, where support is still experimental
- you need a Python-first library rather than a C/C++ performance library

## Facets
- artifact type: library
- maturity: stable
- function: deep-learning, machine-learning, gpu-computing, benchmarking
- domain: deep-learning, machine-learning, gpu-computing, performance
- platform: cpp, windows, cross-platform
- tags: onednn, oneapi, sycl, openmp, tbb, avx512, amx, aarch64, bfloat16, neural-network-primitives, cpu-optimization, linux, macos, gpu

## Member repositories
- uxlfoundation/oneDNN (main) score 99

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:33.530932+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:23:45.491860+00:00, confidence not recorded.
  - readme: https://github.com/uxlfoundation/oneDNN (fetched 2026-08-28T04:08:33.530932+00:00, sha 8691cf6e698b)
  - homepage: http://uxlfoundation.github.io/oneDNN/ (fetched 2026-08-29T09:16:10.216658+00:00, sha 2807002033d9)
- Data as of 2026-08-30T08:39:29.467469+00:00.