Development and verification of Hornet – A CUDA-optimized, Monte Carlo-agnostic, continuous-energy cross-section processing code

Abstract

Background Hornet is introduced as a GPU-optimized nuclear data processing code designed to accelerate neutron cross-section handling and A Compact ENDF (ACE) file sampling for incident neutron data and thermal scattering laws. Hornet supports continuous-energy Monte Carlo applications and is Monte Carlo–agnostic, making it versatile for academic use and easy integration into various Monte Carlo frameworks.

Methods Hornet is implemented in CUDA C++ using an object-oriented design to take full advantage of Graphical Processing Unit (GPU) parallelism. Verification was conducted through both standalone tests and integration with GPU-optimized REActor Physics Monte Carlo (GREAPMC) – an in-house Graphical Processing Unit (GPU) Monte Carlo code – across standard benchmark simulations, including International Atomic Energy Agency (IAEA) INDC (USA)-107 pin-cell and Mosteller benchmarks. Performance and precision were assessed by comparing Hornet’s output against that of Monte Carlo Simulation (MCS), an in-house CPU-based high-fidelity Monte Carlo code.

Results Across all benchmark scenarios, Hornet matched MCS in accuracy, delivering Root Mean Square (RMS) power differences below 0.1% in pin-by-pin power distribution within an OPR1000 fuel assembly. Remarkable tracking rate enhancements were also observed compared to MCS. Hornet’s ability to maintain continuous-energy treatment on GPU hardware was demonstrated with precise modeling of bound and free thermal scattering effects.

Conclusions Hornet enables robust and efficient GPU-based continuous-energy nuclear data processing, producing high-precision results on par with CPU-based Monte Carlo codes. Its object-oriented CUDA C++ architecture ensures maintainability and adaptability. Tested extensively with GREAPMC, Hornet proves to be a powerful tool for modern nuclear analysis and reactor design. Planned enhancements such as support for photo-atomic ACE file processing will broaden its applicability further.