New parallel iterative solver PETSc

Introduction

A new parallel iterative solver PETSc has been implemented on Flux 2021. Available in Beta Mode, PETSc can be selected in Solving Options and will provide a parallel iterative solver. On very large 3D projects (more than 50 000 nodes), the computation time will be reduced by taking advantage of parallel computing. Moreover, it will also significantly reduce memory consumption (compared to MUMPS). As it is based on distributed computing, the list of compute node needs to be set. It can be done through Flux Supervisor options, in parallel computing tab.

About PETSc

PETSc is a suite of parallel solvers for scientific applications. The PETSc solver offered by this new version of Flux is the first parallel iterative solver. The new PETSc solver allows the users to drastically reduce the memory (RAM) consumption during the solving process, like other iterative solvers, while being nicely scalable with supports for MPI. Using this new option allows taking advantages of all CPU resources of any computers, laptops or HPC nodes thanks to its low needs in memory.

Note: Find more information at the following official link https://www.mcs.anl.gov/petsc

How to use PETSc in Flux

For the first version of this solver, Flux automatically selects in the PETSc suite the best solver and preconditioner depending on the project to solve to offer the fastest solving steps. In Beta mode, the PETSc solver can be selected in Solving options by Solving > Solving process options > Edit:



A new dialog box should appear, allowing the selection of the solver PETSc: Parallel iterative solver in the General options of the Linear system solver tab: