Multiphysics Simulation Towards the Exascale: Memories, Tasks, Data, and Stardust

Event Sponsor: 
Argonne Leadership Computing Facility Seminar
Start Date: 
Jul 29 2016 - 10:00am
Building 240/Room 1406-1407
Argonne National Laboratory
Bronson Messer
Speaker(s) Title: 
Oak Ridge National Laboratory
Rick Stevens

The answers to some scientific questions are to be found in phenomena so complex that the only effective way to uncover them is through simulation and subsequent comparison to experiment or observation.

Near-future HPC platforms promise complexity themselves: deep memory hierarchies, massive parallelism, and sophisticated data analysis workflow infrastructures. Stellar astrophysics simulation is a quintessential example where these twin complexities collide: The origin of the elements that make up the world around us (including ourselves), the production of gravitational waves, and the fundamental nature of neutrinos are all issues that require leadership-class stellar astrophysics simulations for progress to be made in their understanding. I will describe how current and future HPC architectures shape choices and decisions for multiphysics software design and what new things about supernovae, neutron stars, and black holes might be uncovered at the exascale if efficient use of computational power can be realized.