Enhancing PWR Monte Carlo Simulations with GREAPMC: A GPU-accelerated Approach

Abstract

This article introduces the novel methods and performance of GREAPMC (GPU-optimized REActor Physics Monte Carlo), a GPU-accelerated Multigroup Monte Carlo (MC) code designed specifically for pressurized water reactor (PWR) simulations. The code incorporates two distinct approaches to optimize history-based neutron tracking for GPUs. The first novel approach involves the dynamic replacement of the dead particles with new ones during the execution of the transport loop. The second innovative technique optimizes efficiency by capping the history length at a predefined maximum number of interactions and then sorting and invoking the kernel with only the surviving neutrons and automatically adjusting the maximum interactions, considering cycle time during the inactive cycle. Both strategies exhibit exceptional acceleration when compared to MCS, a high-fidelity Monte Carlo code developed at UNIST. The most significant acceleration is achieved through the dynamic truncation method. GREAPMC employs a cell-based geometry modeling approach. Simulation results reveal that GREAPMC maintains a consistent execution time irrespective of the number of cells, that are increased by adding more axial regions. This with MCS, where execution time increases as the number of cells grows. Verification of GREAPMC contrasts is conducted through a pin-by-pin flux tally. The tally bin searching scheme demonstrates efficiency, with simulation times remaining nearly constant in GREAPMC, whether or not tallies are employed. This is in stark contrast to MCS, where the introduction of tallies leads to increased simulation time.