Development and Verification of the GPU-Based Monte Carlo Particle Transport Program MagiC

Monte Carlo (MC) methods, due to their strong geometric simulation capabilities, comprehensive physical modeling, and minimal simulation approximation, are widely applied in areas such as radiation transport, physical criticality safety analysis, shielding design, and radiological medicine. With the increase in the scale of high-resolution numerical simulation problems and computational demands, traditional serial Monte Carlo programs can no longer meet the requirements of high-performance computing. In order to enhance the computational performance and efficiency of Monte Carlo programs, parallelization based on Graphics Processing Units (GPUs) has become an important development direction. Traditional Monte Carlo programs typically only utilize Central Processing Unit (CPU) resources for computation, which means users equipped with Graphics Processing Units (GPUs) cannot fully exploit their computational capabilities, resulting in resource wastage. The floating-point computing performance of GPUs on personal computers is significantly superior to CPUs. Therefore, optimizing Monte Carlo programs to fully utilize GPU performance will improve computational efficiency, making the overall computing process more efficient and flexible. This paper develops the Monte Carlo particle transport program MagiC based on GPUs. Building upon traditional parallel algorithms based on history simulation, the paper implements GPU parallel algorithms based on event simulation. Compared to CPU-based Monte Carlo particle transport programs, MagiC can achieve higher flow rates and memory bandwidth. The effectiveness and reliability of MagiC are verified through application cases in reactors and nuclear medicine. The research has certain engineering application value.