Optimizing Next Generation Hydrodynamics Code for Exascale Systems

Authors: Dana Akhmetova (KTH Royal Institute of Technology), Sumathi Lakshmiranganatha (University of Wyoming), Diptajyoti Mukherjee (Allegheny College), Frederick Oullet (University of Florida), Patrick Payne (Los Alamos National Laboratory), Nicholas Stegmeier (South Dakota State University), Christoph Junghans (Los Alamos National Laboratory), Robert Pavel (Los Alamos National Laboratory), Vinay Ramakrishnaiah (Los Alamos National Laboratory)

Abstract: Studying continuum dynamics problems computationally can illuminate complex physical phenomena where experimentation is too costly. However, the models used in studying these phenomena usually require intensive calculations, some of which are beyond even the largest supercomputers to date. Emerging high performance computing (HPC) platforms will likely have varied levels of heterogeneity, making hybrid programming with MPI+X essential for achieving optimal performance. This research investigates hybrid programming and unconventional approaches like machine learning for a next generation hydrodynamics code, FleCSALE, in the context of tabular equation of state (EOS). We demonstrate an overall 5x speedup to the code, the use of GPUs to accelerate EOS tabular interpolation, and a proof of concept machine learning approach to EOS.

Best Poster Finalist (BP): no

Poster: pdf
Poster summary: PDF


Back to Poster Archive Listing