Performance Evaluation of GREAPMC for Continuous Energy Reactor Physics Problems

Abstract

This article investigates the performance of GREAPMC (GPU-optimized REActor Physics Monte Carlo) for continuous-energy nuclear reactor physics problems. GREAPMC leverages several GPU-specific optimizations, such as the use of the float4 construct to access multiple cross-section values in a single operation, caching the total macroscopic cross-section, and utilizing single-precision nuclear data to optimize memory usage and computational speed. The performance of GREAPMC is assessed in comparison to MCS, an in-house, high-fidelity, CPU-optimized MC code, with a focus on tracking rates and computational efficiency for both a fuel assembly and a core problem. The results demonstrate that GREAPMC outperforms MCS, achieving higher tracking rates and greater computational efficiency at lower costs. This highlights the significant advantages of GPU-accelerated MC code, particularly for large-scale nuclear reactor simulations that require high computational power and fast processing times. Future work will focus on further optimizing GREAPMC by implementing energy event-wise particle sorting. Additionally, the adoption of based neutron tracking will be explored.