Parallel performance of the PETSc Krylov methods applied to linear systems of 3D nanodevice simulators

One of the most demanding part of the 3D simulators is the resolution of the linear systems. The parallel performance of solver methods and preconditioning techniques of the latest PETSc numerical library are presented in this work, taking as a reference linear systems obtained from a 3D MOSFET simulation. We have obtained an important scalability with the number of processors when parallel iterative solvers based on Krylov methods are taken into account.

keywords: