Accelerating AZKIND Simulations of Light Water Nuclear Reactor Cores Using PARALUTION on GPU

This paper presents the results of the accelerated solution of the linear algebraic system Av = b arising from a nodal finite element method implemented in the neutron diffusion code AZKIND to solve 3D problems. The numerical solution of full nuclear reactor cores with AZKIND implies the generation of large sparse algebraic systems that produce bottle-necks in the iterative solution. Aiming to alleviate the overload of the algorithm, an acceleration technique has to be implemented. Consequently, a Fortran plug-in of the open source linear algebra PARALUTION library (C ++) was integrated into the AZKIND source code (Fortran 95). This implementation allows AZKIND to use GPUs as well as CPUs, threading into the GPU thousands of arithmetic operations for parallel processing. Selected examples of preliminary investigations performed for a cluster of nuclear fuel assemblies are presented and the obtained results are discussed in this paper.