Beyond Exascale: The Future of HPC

Presenter: Marc Snir, Faiman Professor, Department of Computer Science, University of Illinois at Urbana-Champaign

Date: July 11, 2018

Slides: https://uofi.box.com/s/0vxsc7mo59g8fu3a2maff0s3y8rpbqbu

Video: https://youtu.be/PShwfKLoQOY

Abstract

HPC platforms have evolved in the last 25 years in a predictable manner: They have been designed by assembling commodity compute engines into large clusters. Performance improvements have resulted from the continued improvement in the cost/performance of compute engines and the increasing cost of leading supercomputers. The end of Moore’s law also means an end to this period or smooth evolution. “Conventional HPC” is likely to plateau in performance and slowly improve in cost/performance. Significant performance advances will require larger budgets and more specialized hardware and software.

My talk will discuss the technical and socio-political aspects drivers of this likely evolution.

Target Audience: Those interested in the future of HPC.

Prerequisites: None.

Biography

Marc Snir is Michael Faiman Professor in the Department of Computer Science at the University of Illinois at Urbana-Champaign. He currently pursues research in parallel computing. He was Director of the Mathematics and Computer Science Division at the Argonne National Laboratory from 2011 to 2016 and head of the Computer Science Department at Illinois from 2001 to 2007. Until 2001 he was a senior manager at the IBM T. J. Watson Research Center where he led the Scalable Parallel Systems research group that was responsible for major contributions to the IBM SP scalable parallel system and to the IBM Blue Gene system.

Marc is AAAS Fellow, ACM Fellow and IEEE Fellow. He has Erdos number 2 and is a mathematical descendant of Jacques Salomon Hadamard. He recently won the IEEE Award for Excellence in Scalable Computing and the IEEE Seymour Cray Computer Engineering Award.