Abstract:
Parallel multigrid: effectiveness on modern numerical technologies The goal of the work is to demonstrate the capability of multigrid method on the modern processors architectures. The method is used to numerical solution of 3D parabolic and elliptic differential equations with essential large-scale discontinuous coefficients on non-uniform grids. It is shown that the computer code solves complex problems and keeps effectiveness with increasing processors number on the modern processor architectures: on multi-cores processors of traditional architecture, graphics processors Nvidia and coprocessor Xeon Phi architecture Intel MIC.