cloudparts

Unified Acceleration (UXL) Foundation

The Linux Foundation project that took over Intel’s oneAPI: an open specification and a set of C++ libraries for CPUs, GPUs and other accelerators, steered by chip and cloud companies as a vendor-neutral answer to single-vendor stacks.

1 projects in this catalogPart of the Linux FoundationWebsite

What it is

The Unified Acceleration Foundation, or UXL Foundation, was announced on September 19, 2023 at the Linux Foundation’s Open Source Summit Europe in Bilbao. It is hosted by the Joint Development Foundation, the Linux Foundation’s legal home for specifications, and grew out of Intel’s oneAPI initiative: Intel contributed the oneAPI specification and the open-source projects that implement it. Its founding steering members were Arm, Fujitsu, Google Cloud, Imagination Technologies, Intel, Qualcomm and Samsung; AMD said on October 20, 2026 that it would join as a steering member.

Its aim is portable software for accelerated computing: code written once against open standards, chiefly the oneAPI specification and Khronos’s SYCL, that runs on many vendors’ CPUs, GPUs and accelerators. The foundation hosts six libraries, oneDNN for deep-learning primitives, oneDAL for data analytics, oneMath, oneTBB for CPU parallelism, oneDPL for parallel C++ algorithms and oneCCL for collective communication, plus tools such as the oneAPI Construction Kit for bringing SYCL to new hardware.

For someone choosing technology, the badge means a library’s name, specification and roadmap are governed jointly by several hardware vendors rather than by Intel alone, and that the code is open source. It doesn’t say how well a library runs on a given vendor’s hardware: most of the code still comes from Intel, and that support varies by library and backend. There are no maturity stages.

Its projects in this catalog

oneTBBOpen sourceoneAPI Threading Building Blocks, formerly Intel TBB: a C++ library that splits work into tasks and runs them on a shared pool of worker threads with a work-stealing scheduler, plus parallel algorithms, concurrent containers and a scalable memory allocator.

In depth

How it is organised#

A Steering Committee of the paying steering members governs the foundation and publishes its minutes. Rod Burns of Codeplay, an Intel subsidiary, chaired it from the launch until September 2025, when Manuj Sabharwal of Qualcomm and Andrew Wafaa of Arm took over as co-chairs. Technical coordination of the specifications, the open-source projects and shared infrastructure happens in a Working Group, and Special Interest Groups bring domain experts, users and developers together; in 2026 five were active, for Language, Math, AI and Scientific Computing, Hardware and Memory-Centric Computing, while the Image and Safety-Critical SIGs had been retired.

Joining a SIG or a meeting needs no membership: meetings are on a public calendar, and each group keeps its notes and slides in the foundation’s GitHub repository. Each library keeps its own maintainers and repository under the uxlfoundation GitHub organisation.

Where it came from2019–2026

Intel launched oneAPI in 2019 as one programming model for CPUs, GPUs, FPGAs and other accelerators: Data Parallel C++, based on the Khronos SYCL standard, plus a set of performance libraries, published as an open specification. On September 19, 2023 the Linux Foundation announced the UXL Foundation as “an evolution of the oneAPI initiative”, with Arm, Fujitsu, Google Cloud, Imagination Technologies, Intel, Qualcomm and Samsung taking part, and Intel contributed the specification and its implementations. Rod Burns of Codeplay, who chaired the steering committee, described the aim as “a multi-architecture and multi-vendor programming platform for all accelerators”.

Coverage at the time noted that Nvidia, with its own CUDA programming model, was the most conspicuous absentee, and AMD had not joined either. In September 2025 Burns handed the chair to co-chairs from Qualcomm and Arm. The libraries’ repositories moved from Intel’s oneapi-src GitHub organisation to uxlfoundation. In October 2026 AMD and Intel said they would make oneDNN the default open-source library for x86 acceleration in PyTorch, and that AMD would join the foundation as a steering member.

Why it’s built this way

Build on SYCL instead of a new language
oneAPI’s Data Parallel C++ already built on SYCL, an open Khronos standard for C++ on accelerators, so the foundation started from a standard no single vendor owns rather than an Intel-only language.
Host the specification at the Joint Development Foundation
The JDF gives projects that develop technical specifications and standards a corporate and legal home, staff and an industry network, which suited a group whose main product is the oneAPI specification.
Open SIGs, paid steering seats
Anyone can take part in the technical groups without membership, while steering members fund the foundation and govern it.

Timeline

  1. 2019

    Intel launches oneAPI, with Data Parallel C++ based on SYCL

  2. 2023

    UXL Foundation announced on September 19 in Bilbao with seven steering members

  3. 2025

    Rod Burns steps down as chair in September; Qualcomm and Arm take over as co-chairs

  4. 2026

    AMD says on October 20 it will join as a steering member

Sources · facts checked 2026-10-28

Improve this page on GitHub