American Nuclear Society
Home

Home / Publications / Journals / Nuclear Science and Engineering / Volume 197 / Number 8

High-Performance and High-Fidelity GPU-Based Monte Carlo Solutions to the BEAVRS Benchmark

Kyung Min Kim, Jaeuk Im, Namjae Choi, Han Gyu Lee, Han Gyu Joo

Nuclear Science and Engineering / Volume 197 / Number 8 / August 2023 / Pages 1823-1844

Technical papers from: PHYSOR 2022 / dx.doi.org/10.1080/00295639.2022.2148812

Received:August 19, 2022
Accepted:November 14, 2022
Published:July 7, 2023

The BEAVRS benchmark is solved by PRAGMA, the graphics processing unit (GPU)–based continuous-energy Monte Carlo code. The solutions consist of the detailed simulation results for the two cycles that involve the reactivity and pin power distribution information for the zero-power physics tests and depletion. Primary results at hot zero power, such as the critical boron concentration at various rodded conditions, control rod bank worth, isothermal temperature coefficients, and assemblywise detector signal, are compared with the measured data. Core-follow calculations are performed with varied power, and the resulting boron letdown curves are compared with the measured one. Hot full-power depletion is also performed and the resulting pinwise power distributions of cycle 1 are compared with the nTRACER results. The comparison with the measured data and also with the nTRACER results demonstrates the high solution fidelity of PRAGMA. In all the calculations, PRAGMA uses a tremendously large number of histories, ranging from up to hundreds of millions per cycle, that are used to fully exploit the massive parallel computing capacity of GPUs. The execution time of the entire core-follow calculation with about 30 burnup steps takes less than 16 h on a single rack of computing nodes mounted with 24 gaming GPUs, which represents considerably high Monte Carlo core calculation performance.