GPU Based Two-Level CMFD Accelerating Two-Dimensional MOC Neutron Transport Calculation

The Graphics Processing Units (GPUs) are increasingly becoming the primary computational platform in the scientific fields, due to its cost-effectiveness and massively parallel processing capability. On the other hand, the coarse mesh finite difference (CMFD) method has been one of the most popular techniques to accelerate the neutron transport calculation. The GPU is employed into the method of characteristics (MOC) accelerated by two-level CMFD to solve the neutron transport equation. In this work, the Jacobi method, the successive over-relaxation (SOR) method with red-black ordering, and the preconditioned generalized minimum residual (PGMRES) method are applied to solve the linear system under the framework of CMFD. The performance of these linear system solvers is characterized on both CPU (Central Processing Unit) and GPU. The two-dimensional (2-D) C5G7 benchmark problem and an extended mock quarter-core problem are tested to verify the accuracy and efficiency of the algorithm with double precision, as well as the feasibility of massive parallelization. Numerical results demonstrate that the desired accuracy is maintained. Moreover, the results show that the few-group CMFD acceleration is effective to accelerate the multi-group CMFD calculation. The PGMRES method shows remarkable convergence characteristics compared to the Jacobi and the SOR methods. However, the SOR method shows better performance on GPU for solving the linear system of CMFD calculation, which reaches about 2,400x speedup on GPU with two-level CMFD acceleration compared to the CPU-based MOC calculation.

Abstract licence: CC BY

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.