Invitation for experts to collaborate within a new community forum for accelerated computing
Today we are excited to join a new Community Forum for Accelerated Computing that will build on the work done to define the oneAPI™ specification. Codeplay®* has been active in contributing to oneAPI and now I am honored to be nominated as the first chairperson to lead the expansion of the forum. I look forward to helping drive an increasingly open approach to shape the future of an open specification for accelerated computing.
I invite experts from the community to join as collaborators within the Community Forum for Accelerated Computing to help define and shape the future of an open standard, open source heterogeneous programming platform.
The use of heterogeneous architectures has increased rapidly over the past decade as applications take advantage of highly parallel accelerators capable of bringing huge performance gains to many algorithms and software.
During this time, Codeplay has been defining and implement open standard programming models for the latest cutting-edge processors. The company’s involvement with OpenCL™ goes back many years, and Codeplay has helped to define the original and latest specifications for SYCL™. Open standards are crucial for application developers to ensure that software and libraries that are commonly used around the world can run on current and future architectures.
oneAPI was initiated by Intel® in 2019 with support and contributions from key members of the industry. The initiative represents a huge investment in the development of open source software. oneAPI consists of an open specification alongside largely open source implementations of this specification: The DPC++ compiler based on the SYCL standard exists with libraries for commonly used operations including those for BLAS and neural networks.
Codeplay has been a key contributor to oneAPI, working in partnership with Argonne, Lawrence Berkeley and Oak Ridge National Laboratories in the United States to bring support for additional GPUs to oneAPI. This work is delivering oneAPI and SYCL support to Perlmutter, Polaris and Frontier supercomputers, and with it adding Nvidia® and AMD GPU compatibility for oneAPI and DPC++ (oneAPI’s SYCL implementation). So, it’s exciting for us to take a more integrated role within the evolution of oneAPI and the forming of the “community forum for accelerated computing.” This new structure will facilitate an open forum for the definition of open standard interfaces for the open source projects that are implemented to enable the oneAPI specification.
Our goal is to enable this platform for all developers and all processor targets in the future, providing an open alternative to the walled garden proprietary APIs and programming models.
To achieve this, we need to engage with and collaborate across the industry to build a platform that meets the needs of modern heterogeneous architectures. This includes both hardware companies building the latest accelerators and software developers who are building applications, libraries and frameworks that need the crucial performance delivered by these processors. We need to ensure that the interfaces that are being designed and implemented are fit for purpose to enable the current and next generation of software and accelerators.
The Community Forum for Accelerated Computing
The new era of the Community Forum for Accelerated Computing begins now, and we are setting out a plan to deliver a truly open platform, starting with these steps.
- We are publishing the latest draft specification including new oneDNN Graph definitions
- We will agree on governance and voting rules and publish these
- The working groups have an open membership and nomination approach for leadership positions
- The timetable for working group meetings and nominations will be published
Join us on this new journey towards an open standard, open source heterogeneous future.
Find out about all the current working groups, how to get involved, and provide us with feedback through this web page.* Codeplay is an Intel company