Accelerating cluster dynamics simulation of fission gas behavior in nuclear fuel on deep computing unit–based heterogeneous architecture supercomputer

Abstract

High fidelity simulation of fission gas behavior is able to help us understand and predict the performance of nuclear fuel under different irradiation conditions. Cluster dynamics (CD) is a mesoscale simulation method which is rapidly developed in nuclear fuel research area in recent years, and it can effectively describe the microdynamic behavior of fission gas in nuclear fuel; however, due to the huge cost of computation needed for CD model solution, the application scenario of CD has been limited. Thus, how to design the acceleration algorithm for the given computing resources to improve the computing efficiency and simulation scale has become a key problem of CD simulation. In this work, we present an accelerating cluster dynamics model based on the spatially dependent cluster dynamics model, combined with multi optimization methods on a DCU (deep computing unit)-based heterogeneous architecture supercomputer. The correctness of the model is verified by comparing with experimental data and Xolotl—a software of SciDAC program from the U.S. Department of Energy’s Office of Science. Furthermore, our model implementation has a better computing performance than Xolotl’s GPU version. Our code has gained great strong/weak scaling performance with more than 72.75%/84.07% parallel efficiency on 1024 compute nodes. This work developed a new efficient model for CD simulation of fission gas in nuclear fuel.