Authors: John Linford (ARM Ltd), Jeffrey Young (Georgia Institute of Technology), Roxana Rusitoru (ARM Ltd), Andrew Younge (Sandia National Laboratories), Steve Poole (Los Alamos National Laboratory), Simon McIntosh-Smith (University of Bristol), Pavel Shamis (ARM Ltd), Mitsuhisa Sato (RIKEN Center for Computational Science (R-CCS)), Valerio Schiavoni (University of Neuchâtel, Switzerland), Christelle Piechurski (GENCI, France), Michèle Weiland (Edinburgh Parallel Computing Centre (EPCC)), Jonathan Beard (ARM Ltd)
Abstract: This BoF brings together experts and luminaries from the Arm HPC Users Group to share experiences and lessons learned from developing, procuring, hosting, maintaining, using, and expanding Arm-based HPC and hyperscale systems over the last decade. RIKEN’s “Fugaku” has held the #1 spot on the last three Top500 supercomputer lists, Arm enables affordable HPC in the cloud with AWS Graviton 2 and Ampere Altra, and Bristol’s “Isambard” was recently upgraded to the largest Arm-based HPC system in Europe. We will expand on the recent announcements of NVIDIA Grace, ETRI K-AB21, and SiPearl Rhea with a look at future Arm-based systems.
Long Description: Arm availability and usability has grown substantially in the HPC community as demonstrated by Arm-based projects around the world. The Japanese Post-K created “Fugaku”, the #1 system on the last three Top500 lists. The European Mont-Blanc, U.S. DOE/NNSA Vanguard, and the UK’s GW4/EPSRC Isambard projects continue to generate strong proof points for Arm in HPC. The commercial viability of Arm in HPC is demonstrated by NVIDIA's official support of Arm as a GPU host in all their HPC products and the availability of Arm-based servers from HPC vendors. Future extreme scale systems based on NVIDIA Grace, SiPearl Rhea, and ETRI K-AB21 are just over the horizon. This BoF, hosted by the Arm HPC User Group (A-HUG), will be the fifth SC BoF to bring together experts and novices to present visions of the future based on lessons learned over the last decade. Short talks will be offered by ecosystem experts, researchers, and users in various fields and cover a wide array of topics surrounding the Arm ecosystem including hardware architectures, compilers, operating systems, runtime environments, fabric enablement, performance tools, scientific libraries, system software, benchmarks and HPC applications. Presenters will also share roadmap updates and performance results of direct interest to the expected attendees. In particular, we look forward to discussing the lessons learned from designing, developing, procuring, maintaining, using, and upgrading RIKEN’s Arm-based “Fugaku” supercomputer. This landmark system has been in production for over a year and has generated a wealth of experience and understanding. Following initial presentations, we will hold a panel session for active discussion between the BoF participants and audience. Interactive presentation software (Mentimeter) will be used to facilitate virtual panel/audience discussion. The BoF organizers have used Mentimeter for similar events and have received positive attendee feedback on this format. This panel will give attendees the opportunity to engage and interact with early adopters and contributors, learn about current development efforts, and engage the discussion of future trends of Arm-based technologies as the industry advances. Topics will include system architecture, networking, memory system designs, reliability and resilience, and accelerators, all of which are key drivers for the future of the HPC community. The moderator will engage the panel to discuss advances in OS and compilers, runtime components, system libraries, scientific tools, programming models, and application porting experiences. Finally, the panelists will each present a key “lesson learned” and make a prediction based on their experiences with Arm-based HPC. BoF attendees will receive a holistic picture of the current advancements and future trends in both hardware and software of Arm-based supercomputing, with emphasis on practical information and key experiences from the field. After the BoF, we will produce a report of the topics discussed which relate to the Arm HPC ecosystem and send updates to the Arm user mailing list. The Arm HPC Users Group has hosted dozens of events similar to this BoF, including BoFs and workshops at SC18, SC19, and SC20.
Back to Birds of a Feather Archive Listing