학술논문
Parallel $SP_N$ on multi-core CPUs and many-core GPUs.
Document Type
Journal
Author
Kirschenmann, W. (F-EDF-RD) AMS Author Profile; Plagne, L. (F-EDF-RD) AMS Author Profile; Ponçot, A. (F-EDF-RD) AMS Author Profile; Vialle, S. AMS Author Profile
Source
Subject
82 Statistical mechanics, structure of matter -- 82C Time-dependent statistical mechanics
82C70Transport processes
82C70
Language
English
Abstract
Summary: ``This paper presents two parallel Simplified $P_N$ $(SP_N)$ solverimplementations for both multi-core Central Processing Units (CPU) andGraphics Processing Units (GPU). For a nuclear operator such asÉlectricité de France (EDF), the time required to carry out nuclearreactor core simulations is rather critical when dealing withproduction constraints. The $SP_N$ method provides a convenient tradeoffbetween accuracy and numerical complexity and is used in severalindustrial simulations. The parallelization of the $SP_N$ algorithmreduces its computation time. To solve the problem on distributedmemory machines such as PC clusters, Domain Decomposition Methods havebeen investigated. Complementary to this approach, this work aims touse emerging massively parallel processors such as the GPUs as well ascurrent multi-core CPUs. Based on a fine grained parallelism, thissolution achieves good performances on desktop machines. Our multi-coreCPU and many-core GPU implementations allow us to solve 3D $SP_N$problems, respectively, 10 and 36 times faster than our sequential CPUreference.''