Research On The GPU Technology Application In The Reactor Design

With the development of the current third generation reactor system concept, itscore structure is becoming much more complicated and the material is more various,which improves the requirements for reactor physics calculation methods. Becausethe reactor software application takes a lot of time, parallel processing of thetraditional methods cannot meet the needs of the reactor calculation program. Themethod which uses the CPU+GPU heterogeneous parallel mode of general-purposecomputing technology could solve the problem very well. Nowdays, it has beenwidely used and got a lot of achievements in some relevant reactor calculationsoftwares such as ANSYS, GENT4, MCNP etc.In recent years, with the improvement of technology and the production levels,the GPU has got a rapid development in the field of the high performance computingdue to powerful computing ability. For the improvement of the GPU computingpower and the development of the visit to save bandwidth, general purpose computingon GPU is gradually becoming a research hot spot. The GPU utilizes theheterogeneous of CPU+GPU, and the CPU and GPU works together in thisframework. The GPU has been successfully used in the astronomical calculation,biotechnology, medical education, earthquake prediction, etc. Therefore, the researchbased on CPU+GPU heterogeneous has a certain significance.The paper does the research work on the TMSR super calculate platform, and theplatform cluster mainly uses the mainstream Intel CPU and Nvidia GPU. Theresearch content mainly includes the GPU environment building, HPL test, GPUapplication example analysis, etc. Firstly, the paper introduces the modern technologyof programmable graphics hardware, analyzes in detail the whole GPU developmentstage and graphics hardware related technologies and programmable graphicshardware structure, also it points out its internal use of computing resources. Secondly,we also study the theory of CUDA technology, including the CUDA architecture,parallel threads organization, storage model and memory allocation, NVCC compilers,CUDA accelerating library, etc. According to these theoretical basis, we wouldestablish the platform of GPU computing environmment based on the TMSR supercalculation platform. It’s verified that the GPU computing environment can worknormally by HPL test. Finally, we carry out some research about the application ofGPU on the reactor neutron physics analysis software. After a brief introduction ofthe program of software architecture, installation and operation, calculation process,we would analysis the program source code by the analysis tool-UNDERSTAND, andget the project design about the parallel computing. This article explains in detailabout two modules of three-dimensional diffusion calculation and the matrixmultiplication, then it verifies that the design platform of TMSR on GPU computingenvironment could work normally, and part of the calculation speed has improvedsignificantly after the GPU transplantation, and it provides certain reference value andtechnical support for the future study on the GPU computing in the field of reactor.