Massively Parallel Solving of 3D Simplified P_N Equations on Graphics Processing Units

摘要

This paper presents an efficient parallel implementation on Graphics Processing Units (GPUs) for the Simplified P_N (SP_N) calculations in the 3D case. For a nuclear operator such as EDF, the time required to compute nuclear reactor core simulations is rather critical. The SP_N method provides a convenient trade-off between accuracy and numerical complexity and is used in several industrial simulations. The parallelization of the algorithm should allow to reduce the computation time required to solve the eigenvalue problem. To solve the problem on distributed memory machines such as PC clusters, Domain Decomposition Methods have been investigated. Complementary to this approach, this work aims at using emerging massively parallel processors such as the GPUs. Based on a fine grained parallelism, this solution offers the opportunity to achieve good performances at very low cost. Indeed, GPUs provide a large computational power and some specific optimizations allow to near the hardware limits. Our GPU implementation solves 3D SP_N problems 30 times faster than its sequential CPU counterpart.