Computational Math and Statistics Seminar by John R. Tramm - Monte Carlo at Exascale: Adapting Stochastic Particle Transport Methods to Advanced Computer Architectures

Speaker: John R. Tramm, Argonne National Laboratory

Title: Monte Carlo at Exascale: Adapting Stochastic Particle Transport Methods to Advanced Computer Architectures

Abstract:

In this talk, I will discuss the ongoing trend towards stochastic methods in the field of neutral particle transport. Several different stochastic methods, including Monte Carlo and The Random Ray Method, will be introduced along with their advantages over deterministic methods in the field. I will then discuss recent efforts to implement these methods on GPU-based exascale supercomputers -- including the Frontier and Aurora supercomputers, the two largest in the world. Unique optimization challenges stemming from these methods’ stochastic natures will be discussed along with solutions. Additionally, optimizations for Monte Carlo on new AI accelerators (such as the Cerebras Wafer-Scale Engine), and laptop GPUs (such as the Apple M1 Max) will be presented.

Computational Mathematics and Statistics Seminar