Minisymposium Presentation
Ray Tracing Methods for GPU-Accelerated Particle Transport Simulations
Presenter
Dr. Elliott Biondo is an R&D staff member in the HPC Methods for Nuclear Applications Group at Oak Ridge National Laboratory. He received a Ph.D. in nuclear engineering from the University of Wisconsin in 2016. His research involves the development of radiation transport methods for fission and fusion applications. Specific research areas include GPU computing, computational geometry, reactor physics, and Monte Carlo (MC) variance reduction. He is a developer of the Shift and Celeritas MC particle transport codes.
Description
Monte Carlo (MC) particle transport simulations are integral to (1) the high-energy physics event reconstruction process, and (2) estimating radiological quantities within nuclear reactors. This method involves tracking individual particles through a computational geometry using a random walk technique. In practice, this requires interweaving ray tracing operations with computationally expensive physics calculations. The Oak Ridge Advanced Nested Geometry Engine (ORANGE) is a state-of-the-art computational geometry library shared by the Celeritas MC high-energy physics code and the Shift MC nuclear reactor simulation code. This presentation will focus on the ray tracing methods within ORANGE that facilitate efficient execution on the GPUs of leadership-class supercomputers. These methods include reordering operations to leverage single instruction, multiple threads (SIMT) parallelism, hierarchical acceleration structure algorithms, and algorithms that exploit the structured configurations of physical systems. Results from production-level simulations on the Frontier supercomputer demonstrate the effectiveness of these methods.