Science on the Biggest Computers: The Next Generation

Timothy J. Williams
Seminar

In 2018, the Argonne Leadership Computing Facility (ALCF) will stand up one of the largest and fastest new supercomputers in the world. This system, Aurora, will have a unique new architecture and an eighteen-fold improvement in performance compared to today's system. Our sponsors in Washington frown upon it if we switch on a system like this and nobody knows how to use it yet.

The Early Science Program (ESP) gives selected projects dedicated access to the new hardware, before all other users. For a couple of years in advance of the machine, we give these projects training, expert support, and earliest-possible hardware access to develop their application codes for the new architecture. The call for proposals for the Aurora ESP comes in mid-2016.

In this talk you will hear about Aurora, and a precursor system coming in 2016 named Theta. I will survey some current projects targeting Theta, and how they are programming their way toward it. These span a range of scientific areas including brain tissue modeling, cosmology with hydrodynamics, wind turbine simulation, functional materials science, direct numerical simulation of combustion, and atomistic simulation of membrane transport proteins.

Speaker bio:

Tim Williams is a computational scientist at the Argonne Leadership Computing Facility (ALCF), where he serves as Deputy Director of Science. He is manager of the Early Science Program, which prepares scientific applications for early use of the facility's next-generation supercomputers. He works with some of the large-scale projects using ALCF's current IBM Blue Gene/Q supercomputer, especially those in the area of plasma physics. His physics PhD research was in statistical turbulence theory, and he conducted subsequent research in plasma physics, groundwater contaminant flow simulation, and climate modeling—with a nine-year interlude writing pricing and risk software in the financial industry. He has focused on high-performance parallel computing since 1989 at LLNL, NERSC, LANL, and now ANL.