437174 Computer Simulations in the Massively Parallel Era and Beyond

Wednesday, November 11, 2015: 2:00 PM
255A (Salt Palace Convention Center)
Joshua A. Anderson, Department of Chemical Engineering, University of Michigan, Ann Arbor, MI

Computer simulations are now crucial in every field of science. Researchers need flexible simulation, analysis, and visualization tools to complete their work unencumbered. Researchers also need fast tools to reduce the design, experiment, iterate latency. Single threaded performance hit a brick wall many years ago, so researchers can no longer just run old code on new hardware to run simulations faster. Moore's law still gives us double the transistors with each new generation (for now), and all those go into wider parallelism. So today, and for the foreseeable future, all codes must go parallel or be left behind. In this talk, I will give a perspective on massively parallel computing in the soft matter field - where it has been and where it is heading. And I will highlight my own research results where high performance computing was critical. I lead the development of HOOMD-blue, a general purpose particle simulation toolkit that supports molecular dynamics of point and anisotropic particles, and hard particle Monte Carlo. It uses massively parallel GPUs to execute fast, and scales to all 18,000 GPUs on the Titan supercomputer. I value usability just as much as performance, and will discuss how tools with a Python language interface allow researchers to quickly and easily develop complex workflows and combine simulation, visualization and analysis into a single cohesive whole.

Extended Abstract: File Not Uploaded