Source: Own Illustration
Recently, we had the honour of experiencing a remarkable talk by Bronis R. de Supinski, the CTO of Livermore Computing, who dove deep into the design and insights behind the NNSA (National Nuclear Security Administration, part of the United States Department of Energy) Exascale System, El Capitan.
The Architecture of the NNSA Exascale System
The event highlighted the groundbreaking launch of the first U.S. exascale system focused on national security, initiated by Livermore Comnputing (LC) at Lawrence Livermore National Laboratory (LLNL) in collaboration with Hewlett Packard Enterprise (HPE). A central focus was on the discussion of the AMD MI300A, the first Accelerated Processing Unit (APU) for data centres. APUs combine a normal processor with additional coprocessors on one chip. The coprocessors can perform specific calculations (e.g., matrix-matrix-multiplications) significantly faster than a classic processor. Other pioneering innovations include HPE’s Rabbit modules, which are integrated into the system as fast local data storage.
Exascale computing refers to high-performance computers that can perform at least one ExaFLOP (unit of measurement for the ability of a computer system to perform one trillion floating point operations per second 10^18 FLOPS). These systems are seen as the next evolutionary step in computing power and have the potential to tackle complex scientific, technological, and industrial challenges by performing a high number of calculations on large amounts of data in a very short time. With 2 ExaFLOPS, El Capitan is expected to replace the previous fastest supercomputer in the world, Frontier (1.1 ExaFLOPS), when it becomes operational.
During the event, Bronis R. de Supinski gave a comprehensive overview of the architecture of El Capitan (high-performance supercomputing system developed as the first US exascale system for national security) and provided valuable updates on the status of the installation. In particular, the presentation emphasised energy efficiency considerations in large-scale systems and showed how El Capitan’s innovations are targeting significant improvements in energy efficiency.
Strategies for Large Computing Systems
Bronis R. de Supinski, as CTO, was instrumental in formulating and implementing the strategy for large-scale computing at LLNL. His expertise and insights enriched the event and provided a glimpse into the future of high-performance computing. His dual role as Professor of Exascale Computing at Queen’s University of Belfast further emphasised the depth of expertise shared during the talk.
The event took place on 29 November and attracted a diverse audience interested in understanding the intricacies and advances in the field of exascale computing. For those who could not attend in person, the ability to participate remotely via Zoom provided a seamless experience to engage with the content of the event.
Thank you to everyone who attended this outstanding talk. We look forward to more fascinating events in the future!
Responsible for the content of this article are Simon Schwitanski and Malak Mostafa.
Leave a Reply