Current Efforts with ExaSMR—a Monte Carlo Radiation Transport Application

The ExaSMR project is aimed at improving the ability to accurately predict the operating behavior of small modular nuclear reactors to ensure their operation will be safe, reliable, and economical.

Early access to the Summit supercomputer at Oak Ridge National Laboratory (ORNL) suggests 30x–40x performance improvement is possible on a Monte Carlo radiation transport application.

Preparing Applications for an Exascale Ecosystem

The major goal of the US Department of Energy’s (DOE’s) Exascale Computing Project (ECP) is to ensure targeted science applications are exascale ready and poised to address their challenge problems. To quantify improvements over time, ECP Application Development teams develop and perform baseline calculations measurable on both current machines and future platforms as they become available.

Current Efforts with ExaSMR—a Monte Carlo Radiation Transport Application

Monte Carlo transport is used to simulate how radiation particles disburse throughout a physical system such as a nuclear reactor. In an operating reactor, this particle distribution determines where and how much heat is generated and is key to ensuring safe and efficient operation.

The ExaSMR project recently performed calculations using its Monte Carlo radiation transport application on the Summit supercomputer at ORNL. This effort also quantified baseline performance on ORNL’s existing Titan supercomputer. Preliminary estimates suggest that large simulations will execute 30x–40x faster on the full Summit machine relative to Titan.

Early Results and Next Steps

simulation of small modular reactor core

Simulations performed using the Shift radiation transport application show how neutrons distribute throughout a depleted small modular reactor core. Courtesy: the ExaSMR project

By increasing computational performance, ExaSMR can simulate a larger number of particle histories, allowing the team to look at the behavior of an operational nuclear reactor with greater detail and accuracy than they could previously. Increased accuracy means they can make comparisons with measured data to validate the software, resulting in gre­­ater confidence in their ability to perform predictive simulations of nuclear reactors. More generally, these recent results demonstrate that the supercomputing landscape is actually quite favorable for science applications, even those that haven’t traditionally achieved high efficiency on supercomputing platforms. The projected performance improvement of 30x–40x for execution on the Summit supercomputer relative to Titan is actually larger than the corresponding 7x–10x increase in peak theoretical performance. This discovery reveals that not only will the team be able to run its codes on the next generation of supercomputers but also can anticipate far greater efficiency than on the previous generation.

Collaboration and Teamwork

This recent activity of ExaSMR relied heavily on collaboration between researchers at ORNL and Argonne National Laboratory (ANL). Although ORNL and ANL currently have supercomputers based on different computing architectures, many commonalities exist between them. Furthermore, as is often the case, ideas that lead to increased performance on one type of computer can be adapted to provide improvements on other types. By maintaining a continuous line of communication between researchers at the collaborating DOE labs, ExaSMR is able to keep its capabilities at the forefront of its field.