Mark Snir, Mathematics and Computer Science Division, Argonne National Laboratory
High-Performance Computing: Exascale and Beyond
Argonne Physics Division Colloquium - 12 Sep 2014
11:00 AM, Building 203 auditorium

Europe, US, China and Japan are making large investments in the race to exascale performance; these investments, and the product roadmaps of vendors, all but guarantee that exascale performance will be achieved early in the next decade, if not sooner. However, as CMOS technology is reaching its limits, it becomes increasingly harder and more expensive to continue increasing supercomputer performance.

Exascale is not the end goal of the high-performance computing enterprise. This enterprise has been a multi-decadal effort to provide increasingly powerful platforms for numerical simulation in science and engineering. Each performance improvement has led to new scientific breakthroughs and the need for continued improvements does not cease with exascale. Can we continue to satisfy this need in the coming decades?

The talk will examine the implications of technology trends on the future of high-performance computing at exascale and beyond. While zettascale is a feasible goal, significant technological discontinuities seem unavoidable: Continuing the trend of performance improvement may require significant changes in underlying technologies, architectures, and software, as well as significant changes in the economic ecosystem that supports today's high-performance computing.

Argonne Physics Division Colloquium Schedule