At Intel InnovatiON, Pat Gelsinger and team outlined Intel’s Super Powered vision for developers where open approaches, choice, and trust are the foundation in our commitments. These are simple words, but the impact of these 3 words are profound when applied to modern computing. By openness, we mean to deliver productivity and support innovation in an open, collaborative way that benefits the entire ecosystem. Choice is the opportunity to choose the right platform that works best for your software and particular solution, and not have that software – like a proprietary language - choose the platform for you. And our actions need to build and merit trust not only for today but in the future – so that the investments that developers make in Intel technologies continue to add value in next generations.
These same principles have applied to oneAPI since its inception several years ago. Pervasive computing was built on standards in language, communication, mechanical design and so much more. Prior to oneAPI though, accelerated computing has for the most part been dominated by hardware-specific programming models that inhibit growth and societal benefits. oneAPI is a bold footprint offering an open, multivendor, multiarchitecture, performant, and productive foundation worthy of industry and developer trust.
Since the oneAPI spec and Intel’s oneAPI tools release last year, the industry has come a long way. Intel has consistently increased the feature richness and performance of oneAPI. I’m excited to see adoption of oneAPI and its elements by Fujitsu and Huawei, as well as academic, government and industry use on Nvidia and AMD GPUs. We’re off to a very good start.
Bold visions become reality through effort – research, education, experimentation, and delivery. As mentioned at Intel InnovatiON, 11 additional institutions joined the initiative as Intel oneAPI Centers of Excellence. Their missions range from strategic application development to training and curriculum and new technologies and services. Together these new centers contribute yet another step from vision to reality for open accelerated computing.
Last year at a oneAPI Developer Summit, Professor Erik Lindahl the principal investigator (PI) for one of the initial oneAPI Centers of Excellence gave a wonderful and candid update of GROMACS’ migration experience highlighting that with oneAPI, “Open is open” in more than words. Among the new CoEs, we have work enabling oneAPI in the open for cross-architecture exascale computing. Another is building standard cross-XPU math libraries. Both help enable a future of choice for future extreme scale scientific research. The 11 centers span technical computing, education, artificial intelligence, and visualization.
Here is a quick look at some of the work being done by the new centers:
To accelerate HPC, 4 new oneAPI Centers of Excellence are leading the way. Oak Ridge National Laboratory (known for its AMD and Nvidia implementations), is advancing the open source HeCBench repository. It contains dozens of benchmark codes to compare different heterogeneous programming approaches including Data Parallel C++, SYCL, HIP, CUDA and OpenMP to migrate to oneAPI. Oak Ridge will study how these samples interact with the underlying hardware. Durham University will research task-based programming using oneAPI to extend the hyperbolic PDE engine ExaHyPE into a OneExaHyPE code that scales across a wide variety of GPU-accelerated machines. The algorithmic and methodological insights benefit many simulation codes.
HPC & Visualization
University of Tennessee at Knoxville will port the open source HPC Ginkgo library to oneAPI and develop high-performance linear algebra components that are integrated in HPC projects to accelerate computing on multiple architectures. For visualization, the center will provide high-end visualization as a service that can be instantly available and universally accessible through Visual Cloud Instances (VCI).
University of California at Berkeley is creating a Center for Energy Efficient Deep Learning focusing on producing energy-efficient algorithms and implementations for deep learning’s most computationally intensive workloads. This work will significantly ease the development of portable implementations across multiple types of architectures.
The following organizations are expanding current Intel Graphics Visualization Institutes of Xellence as oneAPI Centers of Excellence. They are chartering important, unique advancements in graphics, in-situ visualization, and other visual compute technologies across heterogeneous architectures (CPU, GPUs including upcoming Intel Xe GPUs, and other accelerators). While each deserves significant merit – I’ll call out just a few for brevity.
Charles University will innovate sky model lighting simulations and integrate 3D and AI technologies for multiple usages. University of Cambridge will advance cosmological research and open-source code development through its Stephen Hawking Centre of Theoretical Cosmology. University of Texas at Austin will power extreme-scale remote visual analysis for scientific simulations. This will enable flexible, high-fidelity, high-performance interactive analysis across Texas Advanced Computing Center (TACC) computing platforms without having to implement device-specific routines for each system. University of Utah will improve large-scale simulations, data analytics and visualization for scientific workflows. The other new visualization oneAPI Centers of Excellence include Northern Illinois University, University of California at Davis, University of Stuttgart.
Like Intel and like oneAPI, these new Centers of Excellence are “Open!” Together, we’re off to a very good start.