The SuperComputing conference is the largest of its kind, where researchers and the industry come together from the world of high-performance computing. Recently, however, there has been a distinct change in the technologies being discussed at the conference. AI products and services are currently dominating the headlines in the technology space, which was also reflected in the SC23 conference exhibition. This is a testament to the advancements that have recently been made in machine learning and deep neural networks, with ChatGPT and other LLMs gaining lots of attention. Applications like ChatGPT are one of the main reasons that AI was such a big topic at SC23. It can produce compelling information because of the power delivered by accelerators such as GPUs. These accelerators enable AI frameworks to run deep learning and inference at speeds much faster than would have ever been possible using only CPUs.
What is clear is that researchers in the high-performance computing community are embracing AI techniques and algorithms within their existing domains. They are also using accelerators to bring vast speed ups to their traditional algorithms. Large centers such as CERN, Argonne National Laboratory and NASA are taking advantage of GPUs to advance their research faster than they thought was previously possible.
Back to the SC23 conference – I was there to educate on the benefits of using open standards and open source software to build the next generation of AI and HPC software, in particular in my capacity as the chairperson for the UXL Foundation.
I organized a full day tutorial providing an introduction to developing with SYCL, an open standard defined by the Khronos Group. SYCL is an interface for developing software that can be used to develop parallel applications that can run across multiple vendors and architectures, in particular for GPUs but also other accelerators. The tutorial attracted a large group with a mix of researchers and industry attendees and with this being my 7th year of running SYCL tutorials I was impressed and interested by the questions that were being asked. The SYCL standard, like C++, is a crucial dependency on the projects we are developing at the UXL Foundation, so it is important to understand how it will be used.
During the SYCL Birds of a Feather session, I was lucky enough to be invited to present on the UXL Foundation to the SYCL community at SC23 and sit on the panel. The relationship the foundation has with both the SYCL group within Khronos and the developer community, is really important and we want to help broaden the adoption of the SYCL standard. It was interesting to hear the feedback and questions from the attendees, and I look forward to using this to improve and evolve both the UXL Foundation and SYCL to be everything developers need.
One of the most exciting parts of the conference was when some of the UXL Foundation Steering Members were able to get together for our first face to face meeting. We had good representation and were able to talk more about how we can bring success to the UXL Foundation members and community.
During the rest of the week, I presented on the UXL Foundation on the exhibition show floor, helping attendees to understand our mission and how they can help us to achieve it. There is a great interest in the foundation, and we are excited to bring new members into the Working Groups. During the conference there were also some great conversations about how to provide the foundation with the infrastructure it needs to be able to test and integrate software across multiple vendors and operating systems. This will be a real boost to our community and ability to build and test the projects with many configurations.
SC23 is a unique opportunity to meet with so many people from the HPC (and now AI) community and I feel lucky to have been able to talk with so many smart and influential individuals.
Hosted by the Linux Foundation’s Joint Development Foundation (JDF), the Unified Acceleration Foundation brings together ecosystem participants to establish an open standard for developing applications that deliver performance across a wide range of architectures. It is an evolution of the oneAPI initiative and focuses on the development of a specification and open source projects through Working Groups and Special Interest Groups (SIGs).