| Conceptual design of the Generation IV reactor systems, such as SCWR, VHTR,SFR, LFR, GFR, improves the requirements of calculation method and computinghardware for complex core structure, various material and spectrum. Parallel computinghas been widely used, isomorphism CPU achieves good results. But Moore’s Lawencountered bottleneck, the frequency and number of CPU core is restricted for thesharp rise of power consumption, heat dissipation and the limit of transistorsize.Therefore the emerging heterogeneous computing methods develop rapidly andreceive extensive attention.This paper studies the application and features of heterogeneous platforms used innumerical calculation of reactor physics through practical examples. Firstly, we designthe FPGA diffusion calculation module which is implemented and verified on theCyclone IV E FPGA chip. When the frid is divided into30segments, we obtainacceleration ratio2.56. Then we contrast the acceleration of different heterogeneousplatform by solving the PWR benchmark problems, and analysis the effect ofCPU-GPU heterogeneous platform.Both the CPU-FPGA architecture and the CPU-GPU architecture have itsadvantages and limitations. In scientific calculation FPGA module is closer to theunderlying hardware; the optimization of the algorithm is accurate to the clock cycle.But the low frequency and the limited space restrict the scale of problems. GPUarchitecture is more developed; the development environment is friendlier.Programmers can ignore the underlying implementation. GPU is the typical SIMDmode; fixed hardware is subject to specific issues; GPU has significantinternaloverheads which slow it down.FPGA and GPU are both in high-speed development, with the hardwareperformance increasing, designers turn their attention to the field of scientificcalculation. Heterogeneous computing has become an important direction.Heterogeneous plarform can achieve good results, and is of great significance for thedevelopment of reactor physics. |