Intel and Heidelberg University Computing Center (URZ) in the present day introduced that they’ve established oneAPI Academic Center of Excellence (CoE) at UZR. The newly established CoE has a aim to additional develop Intel’s oneAPI commonplace and allow it to work on AMD GPUs. This data is a bit stunning, nonetheless, Intel believes that the know-how ought to work on a variety of processors, regardless of the seller. The heterogeneous {hardware} programming is the primary aim right here. In a Twitter thread, an Intel worker specifies that Intel has additionally been working with Arm and NVIDIA to deliver Data-Parallel C++ (DPC++), a core of oneAPI, to these distributors as properly. That ought to deliver this common programming mannequin to each gadget and adapt to each platform, which is a aim of heterogeneous programming – no matter it is advisable program a CPU, GPU, or another ASIC, it’s coated by a single API, particularly oneAPI.
URZ’s work as a oneAPI CoE will add superior DPC++ capabilities into hipSYCL, which helps methods primarily based on AMD GPUs, NVIDIA GPUs, and CPUs. New DPC++ extensions are a part of the SYCL 2020 provisional specification that brings options equivalent to unified shared reminiscence to hipSYCL and the platforms it helps – furthering the promise of oneAPI software assist throughout system architectures and distributors.
