Seminar Series
2025
April 15
Monte Carlo Neutron Transport on Exascale Computers
Steven Hamilton, Oak Ridge National Laboratory
2:00 p.m., Student Lounge inside Farris Engineering Center
Room 1026, Centennial Engineering Center
Zoom: https://unm.zoom.us/j/91510871864
Meeting ID: 915 1087 1864
Passcode: NE_501 (need to sign in to Zoom)
Abstract: The world's largest supercomputers have become increasingly focused on GPU-based computing architectures. This trend is driven by both the higher energy efficiency of GPUs relative to CPU platforms and the growing interest in applications of artificial intelligence. Researchers wishing to leverage high performance computing platforms for challenging problems must adapt their codes to the unique GPU programming paradigm. This talk presents an overview of the process of porting ORNL's Shift Monte Carlo radiation transport solver to execute efficiently GPUs as part of the DOE Exascale Computing Project. Several of the unique challenges encountered in the porting process will be highlighted. A central theme is that strategies known to be efficient on traditional CPU architectures may encounter unexpected challenges when running on GPUs. A few areas requiring particular algorithmic advances include continuous-energy nuclear data, isotopic depletion, flexible computational geometry, and multiphysics coupling. Simulation results on both Nvidia and AMD GPUs will be presented, including a series of multiphysics calculations on ORNL's Frontier supercomputer that was named a finalist for the 2023 ACM Gordon Bell Prize.
Bio: Steven Hamilton is a Senior R&D staff member in the HPC Methods for Nuclear Applications Group at Oak Ridge National Laboratory. He received BS and MS degrees in Nuclear Engineering from Georgia Tech and a PhD in Computational Mathematics from Emory University. He is a developer on both the Shift Monte Carlo radiation transport code and the Denovo deterministic transport solver, both part of ORNL's SCALE nuclear analysis suite. His research interests are in porting and optimizing scientific software for high performance computing platforms as well as the development of linear and nonlinear solvers for radiation transport and fluid flow problems.
