Progress in Development of High-fidelity 3D Neutron Transport Calculation Code for Reactor Core Based on Heterogeneous Architecture

Abstract

To realize the practical application of the numerical reactor technology on the cutting-edge supercomputer with heterogeneous architecture, or the desktop workstation with consumer graphics cards, advanced lattice physics code based on heterogeneous architecture (ALPHA) was developed by the research center of nuclear power simulation at Harbin Engineering University as the highfidelity 3D neutron transport calculation code for reactor core based on heterogeneous architecture. Firstly, a massively parallel 2D method of characteristics (MOC) algorithm on GPU was proposed in ALPHA. Three parallel schemes were studied based on the ray parallelization in MOC solution method, and the performance optimization of MOCs GPU parallel computing kernel was performed. Numerical results demonstrate that the GPUbased 2D MOC parallel algorithm shows comparative accuracy compared with the other similar codes. The GPU shows powerful computing capacity especially with single precision operation, and the 1080Ti GPU achieves 100× speedup compared with the runtime on a single core of i9-7900 CPU. Secondly, the 2D MOC heterogeneous parallel algorithm which employs the MPI+CUDA programming model was proposed and implemented on CPUs/GPUs heterogeneous system. In this algorithm, the spatial domain decomposition technique provides the coarsegrained parallelism with the MPI protocol, while the finegrained parallelism was exploited through ray parallelization on GPU with CUDA environment. The strong scaling efficiency can be improved by overlapping the MPI communication with computation and applying the asynchronous datacopy between GPU and CPU. Thirdly, the resonance calculation model of ALPHA adopts the fine groupsubgroup secondary discrete strategy to treat the complex overlapping selfshielding effect and uses multigroup kernel to accelerate the performance of resonance treatment on GPU systems. Finally, 3D MOCEX heterogeneous parallel algorithm for neutron transport calculation on the CPUs/GPUs heterogeneous system was proposed, and the whole-core high-fidelity neutron transport calculation was performed high-efficiently and stably with the GPU-accelerated CMFD formulation. Numerical results show that the ALPHA has excellent parallel efficiency and scalable performance in the premise of ensuring the calculation accuracy, which is expected to achieve the lightweight and engineering application of neutronics calculation in numerical reactors.

Keywords

• heterogeneous computing
• gpu
• neutron transport