GPU-Accelerated Neutron Monte Carlo Simulation

by Paul Weimer

Thursday 12 Dec 2024, 09:30 → 10:30 Europe/Berlin

E18/ENE Seminar Room 3268 (TUM PH)

Accurate neutron transport simulations are critical for the design, analysis, and safety assessment of nuclear reactor cores. Conventional Monte Carlo methods like Serpent, MCNP and Geant 4, provide a high degree of fidelity and flexibility in modeling these systems, but their immense computational cost has historically limited their practical use, especially for large, full-core problems and automated parameter optimization. Leveraging the parallel processing powers of GPUs, this computational cost can be mitigated. To fully capture the performance offered by GPUs, a full redesign of current Monte Carlo codes is necessary. The developed GPU-accelerated code incorporates CSG based geometry, continuous-energy cross-sections, and models all major physical phenomena. Preliminary benchmarks against established Monte Carlo tools demonstrate close agreement with reference solutions, while offering substantial performance gains. 

This talk will detail the underlying acceleration strategies, general neutron simulation techniques, and benchmarks/validation tests.