A smiling robot looks at the camera.
Enlarge / Machine studying would not even have a face. But when it ever does, we might like that face to look one thing like this.

We have been following Intel’s oneAPI—a man-made intelligence growth platform designed to summary {hardware} away from the duty of creating AI code—with nice curiosity since its launch final November. This week, Intel and the Heidelberg College Computing Heart (URZ) introduced a brand new Tutorial Heart of Excellence (CoE) which can assist and conduct analysis on the oneAPI platform.

The brand new collaboration instantly adopted Intel’s personal announcement of the oneAPI specification reaching 1.0 standing. The 1.0 milestone is critical, because it allows collaborators to concentrate on adapting {hardware} to an ordinary, fastened implementation, worrying in regards to the specification itself shifting quickly beneath their ft.

URZ’s personal announcement of the oneAPI Heart of Excellence begins by reiterating the raison d’etre of oneAPI itself:

URZ will focus its analysis and programming efforts on a elementary high-performance computing (HPC) problem the place fashionable computer systems make the most of various kinds of {hardware} for various calculations. Accelerators, together with graphics processing items (GPUs) and discipline programmable gate arrays (FPGAs), are utilized in mixture with basic compute processors (CPUs). Utilizing various kinds of {hardware} make computer systems very highly effective and supply versatility for a variety of conditions and workloads. {Hardware} heterogeneity, nonetheless, complicates software program growth for these computer systems, particularly when specialised elements from quite a lot of distributors are utilized in tandem.

One main purpose for this complication is that many accelerated compute architectures require their very own programming fashions. Due to this fact, software program builders have to study and use a distinct—and generally proprietary—language for every processing unit in a heterogeneous system, which will increase complexity and limits flexibility.

oneAP’s cross-architecture language Information Parallel C++ (DPC++), based mostly on Khronos Teams’ SYCL normal for heterogeneous programming in C++, overcomes these challenges with its single, unified open growth mannequin for performant and productive heterogeneous programming and cross-vendor assist.

URZ’s collaboration on the oneAPI normal seems to be a superb begin towards fulfilling Intel’s guarantees that oneAPI might be about all {hardware}, not simply Intel’s. The event efforts might be led by Aksel Alpay, who’s particularly trying to deliver AMD (Radeon) GPU assist to oneAPI.

hipSYCL is a part of the wider SYCL hardware-abstraction ecosystem, and it targets vendor-neutral, near-native performance on CPUs and GPUs alike.
Enlarge / hipSYCL is part of the broader SYCL hardware-abstraction ecosystem, and it targets vendor-neutral, near-native efficiency on CPUs and GPUs alike.

To completely perceive this, we have to dive a bit deeper into oneAPI’s underlying applied sciences. OneAPI is constructed on high of Intel’s Information Parallel C++ (DPC++), which is itself constructed atop C++ and Khronos SYCL code abstraction layer requirements. Alpay is himself the founding developer of the hipSYCL implementation of the SYCL normal, and his work already explicitly helps a number of CPUs, Nvidia GPUs through CUDA, and AMD GPUs through ROCm.

The brand new oneAPI Heart of Excellence will make use of Alpay (and others) so as to add DPC++ capabilities and assist into hipSYCL, which already helps Intel and AMD CPUs, Nvidia GPUs, and AMD (Radeon) GPUs natively. URZ’s researchers and builders will, in flip, have entry to “a world community of specialists” together with these at Intel, in addition to different educational and authorities establishments.

Here is what Professor Vincent Heuveline, CIO of Heidelberg College and URZ Managing Director, needed to say:

For a scientific computing heart to have entry to this stage of experience and work collectively on an open normal with companions from across the globe, is a superb prospect.

The college’s work helps deliver in regards to the guarantees Intel has made since day one in all oneAPI being an successfully vendor-neutral, cross-hardware-platform effort—and, most likely not accidentally, break the near-monopoly Nvidia has loved within the GPU accelerated machine studying ecosystem for years. Breaking that de facto monopoly advantages Intel’s personal new and rising Xe Graphics platform, in addition to AMD’s personal Radeon, which has up to now largely been restricted to “conventional” graphics and gaming use.


Please enter your comment!
Please enter your name here