Co-Designing Next Generation Supercomputing Systems

After reaching exascale, we are now entering a new era for high-performance computing systems for scientific discovery. While centralized computing systems will remain critical, they will become a part of a complex solution that will involve edge devices for data acquisition from experiments, the fog between edge and centralized solutions for data movement and preprocessing, and even disaggregated system in the cloud for execution of scientific simulation, augmented with artificial intelligence methods. The implementation of the various semiconductor manufacturing acts around the world will provide an opportunity for governments to explore new architectural concepts that might provide the necessary customization for the new disparate requirements of scientific discovery. However, without a proper co-design approach involving the entire spectrum of stakeholders, from domain scientists, to application developers, software researchers, hardware designers, physical chip designers, and even material scientists, and the ability to leverage a rich ecosystem of openly available co-design and hardware design tools, we would still remain limited by the “valley of death” between idea and implementation, allowing only large companies to develop concepts with broader market appeal but not completely fulfilling the specific needs of scientific discovery. This BoF will explore challenges, opportunities, and perspective of end-to-end co-design for the next generation of high-performance computing systems, identifying existing successful approaches and potential ways that the researchers in our community could become “lead users”.