Development of a GPU-accelerated 3D neutron dynamics code for PB-FHR

Pebble bed fluoride-salt cooled high temperature reactor (PB-FHR) is a kind of novel nuclear energy systems, combining advanced techniques such as coated particle fuel and molten-salt coolant to achieve better safety and economy performance. The compact core of a PB-FHR is geometrically complicated due to the randomly-packed pebble bed with fuel motion in a complex core chamber, which yields a highly detailed core-modeling for accurate neutron dynamics analysis. Three-dimensional fine mesh finite volume method is expected to achieve better accuracy because of its strong geometry adaptability in detailed core-modeling, but is much more time-consuming compared to other methods such as coarse mesh nodal method. A GPU-accelerated 3D fine mesh neutron dynamics code (GAND) is developed, using conjugate gradient method (CG) to solve the time-dependent multi-group neutron diffusion equations in r-z-θ coordinates. The GAND code is verified by a cylindrical reactor benchmark with good agreement, and is preliminarily applied to a PB-FHR core in both static and transient analysis. Speed-up ratio and other performance of GAND code using different equation preconditioners is studied, a best speed-up ratio of 21.65 has been achieved using Neumann polynomials-preconditioned CG.