Intel Hosted oneAPI Hackfest at DiRAC HPC Research Facility

DiRAC researchers learned about and leveraged oneAPI toolkit for multiarchitecture, high-performance computing with SYCL

Authors:

Nikita Shiledarbaxi, Software Product Marketing Engineer, Intel

Rob Mueller-Albrecht, Software Tools Marketing Manager, Intel

The DiRAC* HPC facility in the UK, in collaboration with Intel and Codeplay*, organized a oneAPI-focused hackfest in May-June'24. The event was composed in two parts – virtual oneAPI training sessions for the DiRAC members, and an Intel oneAPI hackathon held at Codeplay centers. The hackfest participants got an opportunity to learn about the Intel® oneAPI Base Toolkit (Base Kit). They harnessed the potential of Base Kit’s component tools and libraries to migrate scientific workloads to SYCL*, achieving code portability across multi-vendor GPU architectures with a common SYCL codebase.

This blog will give you highlights of the DiRAC’s oneAPI hackfest.

What is DiRAC?

DiRAC is a High Performance Supercomputing facility in the UK that provides scientific researchers with advanced computing services. The research focus areas at the DiRAC community include particle physics, astro physics, cosmology, solar system and planetary science, and nuclear physics. The DiRAC facility is funded by the Science and Technology Facilities Council (STFC), a part of the UK Research and Innovation (UKRI).

oneAPI Training Series at DiRAC Hackfest

The DiRAC community conducted 3 half-days remote training sessions (May 21-23, 2024) on the Intel oneAPI Base Toolkit. It was an introductory training series targeted for developers and/or researchers of all experience levels, from beginners to advanced. On average, 18-20 participants were trained each day.

The attendees were educated on a wide range of concepts based on the cross-platform supported SYCL programming framework. They also learned about various component tools and libraries of the Base Kit that support SYCL – examples include Intel® DPC++ Compatibility Tool for automated code migration from CUDA* to C++ with SYCL, Intel oneAPI Math Kernel Library (oneMKL) for optimized math routines, Intel oneAPI Deep Neural Networks (oneDNN) for accelerated deep learning, and Intel oneAPI DPC++ Library (oneDPL) for expedited SYCL kernels on diverse hardware. Moreover, the sessions also involved training about code profiling and performance bottlenecks analysis tools included in the Base Kit, namely Intel® Advisor and Intel® VTune™ Profiler.

oneAPI Hackathon at DiRAC Hackfest

Following the training series, an in-person 3-days Intel oneAPI hackathon ((June 11-13, 2024) was hosted at the offices of Codeplay in Edinburgh, Scottland. The 6 teams that participated in the hackathon focused on leveraging the Intel oneAPI Base Toolkit for developing new codes for diverse GPUs, and re-factoring CPU-specific codes to enable execution on GPU architectures. They ported their workloads to SYCL for distributed computing through GPU-offloading of computationally intensive tasks. Each team was provided with access to a node with four Intel® Data Center GPU Max 1550 on the DAWN [1] supercomputer cluster.

The participants improvised their cutting-edge projects using oneAPI tools and libraries for a variety of tasks such as migrating CUDA code to SYCL and accelerating math routines like Fast Fourier Transform (FFT) using oneMKL’s SYCL API, parallelizing Fortran* code on Intel GPUs, and fixing performance bottlenecks with the help of Intel Advisor and Intel VTune Profiler.

The participants found the code migration process to be smooth and easily adapted to the usage of oneAPI components. With libraries like Intel® MPI, the teams achieved significant performance boost for their workloads. Around 70% of the participating teams also showed willingness to employ oneAPI tools for further optimization of their research project code. With a 100% success rate of code migration to SYCL, 30% of the teams also benchmarked their results attained using oneAPI and SYCL.

Get Started with oneAPI and SYCL for Multiarchitecture Programming

Start exploring the oneAPI toolkits and SYCL framework for accelerated, multiarchitecture programming today! Migrate your workloads to SYCL and add cross-platform parallelism to your applications with oneAPI for high-performance heterogeneous computing.

We encourage you to check out practical application examples of code migration available in the CUDA to SYCL catalog. Also, explore AI, HPC, and Rendering tools in Intel’s oneAPI-powered software portfolio.

Get the Software

Download the Intel oneAPI Base Toolkit. You can also develop, test, and deploy AI, HPC, and edge-computing workloads using the toolkit’s components on the latest hardware available on Intel® Tiber™ Developer Cloud.

Useful Resources

• oneAPI training series: DiRAC community blog
• Hosting the DiRAC Hackathon: Codeplay blog
• Intel oneAPI Base Toolkit offers core multiarchitecture tools
• Essentials of SYCL
• Migrate from CUDA to C++ with SYCL portal

[1] DAWN is the UK’s fastest AI supercomputer jointly built by Intel, University of Cambridge* and Dell Tec