On October 9th and 10th we hosted our first virtual oneAPI DevSummit, bringing together the whole UXL Foundation community for presentations, panels and lightning talks. It was a great opportunity to learn from the experts, and to take part in key discussions on important topics. There was a strong focus on open source and open standards software development for GPUs and other accelerators. We want to demonstrate that you can develop software that achieves performance portability across processors, from CPUs to GPUs and beyond.
My opening “State of the Nation” presentation talked about the problems we are trying to solve in the UXL Foundation, and how we are tackling some of these by bringing together expertise from across the industry. The first year of the UXL Foundation has seen strong growth in membership, participation in our organized groups, and contributions to the open source projects. Find out more by watching the recording of my talk.
The UXL Foundation projects presented updates on their current development, new contributions and plans for the upcoming year. This is a great way to understand how these projects are working, how you can use them and can contribute. Watch the update from the project maintainers.
We had some really fascinating keynote presentations from members of our community and I’d like to highlight these.
Dave Airlie has been contributing to Linux for many years, and is an upstream Linux GPU maintainer working at Red Hat. His presentation gives us some great insight into what it takes to make open source projects a success, and how this can be applied to the UXL Foundation projects.
Performance portability is the ultimate goal for developers – writing code once and deploying this across multiple architectures from different vendors. GE HealthCare has been using SYCL and oneAPI projects to develop some of their next generation software, and Evgeny Drapkin, Chief Engineer, gave some insight into their experiences, highlighting their success towards achieving performance across GPUs from a single code base.
Dave Murray from Imagination Technologies made a fascinating presentation on how they are adopting SYCL and oneAPI as part of their software strategy for customers, making it possible for you to write software that works across a range of processors in embedded environments.
For AI software, a major factor in achieving performance is memory throughput, and there are new innovations being developed to increase the memory capabilities for GPUs and other processors. Engineers from Samsung presented their proposal for oneMCC, a new programming interface for memory centric computing. This is used with their latest PIM technology and they are looking to collaborate with the community to define standard programming interfaces for oneMCC.
The latest supercomputers must have the software required to run the latest AI workloads to complement the vast array of hardware available. The closing keynote for day 1 was presented by Dr. Priyanka Sharma from Fujitsu, giving insight into the next generation Monaka supercomputer and how it will benefit from the UXL Foundation libraries. The oneAPI projects we are developing will help maximize the performance of AI applications on Monaka.
We also saw some great discussion and debate in the two panel sessions. The first saw Eleanor Brash from Imagination Technologies lead a conversation about open standards and open source with representatives from UXL members. There is a strong commitment across our members to embrace open standards but it is not without its challenges. A replay is available here.
The second panel was focused on the use of open standards in the automotive space, with Zain Khawaja from Autoware quizzing the participants on the next generation of automotive software. There are different challenges when deploying software in embedded systems compared to supercomputers or cloud machines, from power management through to safety concerns. Watch the replay of this discussion here.
Here’s a summary of some of the other presentations you can watch:
- How to develop with math library oneMKL on AMD, Intel and Nvidia GPUs
- Yu-Hsiang shows how the Ginkgo project uses SYCL and oneAPI to implement cross architecture linear solvers
- Fujitsu is using oneDAL to accelerate AI workloads for the Monaka supercomputer
- Roofline AI have made use of the Level Zero specification to map their compiler to different devices
- The SYCLDB project aims to enable AI analytics for RISC-V and other architectures using the oneAPI Construction Kit
I am really pleased with the participation and engagement of the community during the UXL Foundation hosted DevSummit with sessions featuring all our Steering members, and great presentations from the community. If you have not done so already please join our Slack workspace, sign up to the mailing lists and subscribe to our regular updates.
UXL Foundation members from Arm, Fujitsu, Intel and Samsung are presenting and exhibiting at SC24, the international conference for high performance computing, networking, storage, and analysis in Atlanta, GA from November 17-22, 2024.
You can find out more details of UXL member Intel’s activities @ SC24. Please register for the UXL Foundation breakfast meetup at SC24 https://linuxfoundation.regfox.com/uxl-foundation-at-sc-24
#UXLDevSummit2024 #oneAPI #UXLFoundation #OpenSource #AI
The post Hear from the experts – re-watch the UXL Foundation hosted oneAPI DevSummit appeared first on oneAPI.