Development of a GPU-based three-dimensional neutron transport code.

• A GPU-based 3D neutron transport code is newly developed. • Calculating the modified total source on-the-fly to significantly reduce the GPU memory requirement. • An azimuthal angle decomposition algorithm is proposed to reduce GPU memory consumption further. • Geometric and physical information are written into a VTK file for visualization. A GPU-based three-dimensional (3D) neutron transport code, named ThorMOC, is developed in this work. A GPU-parallelized transport sweeping algorithm based on a planar method of characteristics (MOC) is implemented in ThorMOC and a two-level coarse mesh finite difference (CMFD) method is used to accelerate the convergence of the MOC transport solution. To reduce the memory burden of 3D calculation, the transport sweeping algorithm is improved by azimuthal angle decomposition. This developed ThorMOC code is verified by a Takeda benchmark and a C5G7 3D extension benchmark, and the performance of the azimuthal angle decomposition algorithm is evaluated by the whole core calculation of a C5G7 hexagonal benchmark. The numerical results obtained from ThorMOC show a good agreement with those from the Monte Carlo code OpenMC. Besides, the azimuthal angle decomposition algorithm can reduce the GPU storage requirement by more than 40% for the C5G7 hexagonal benchmark. [ABSTRACT FROM AUTHOR]