What is my type of finite element problem?
Introduction
From a simple 2D case to a full 3D model representing a complex electromagnetic device, the finite element problems may be drastically different and does not require always the same needs in terms of computation resources.
- Should I use parametric distribution to decrease the solving time?
- Should I use the solver parallelization to decrease solving time?
For Flux 2D and Flux Skew, the recommended linear solver is Direct (MUMPS) for all the configurations.
Mesh size | Number of parameters in the scenario | Use of parametric distribution | Use of solver parallelization | |
---|---|---|---|---|
Flux 2D | 0-100k | 0-10 | No | No |
0-100k | 10 + | Yes, single machine or cluster | Yes, see recommandations | |
100k + | 0-10 | No | No | |
100k + | 10 + | Yes, single machine or cluster | Yes, see recommandations |
Mesh size (extruded geometry) |
Use of solver parallelization | |
---|---|---|
Flux Skew | 0-300k | No |
300k + | Yes, see recommandations |
Mesh size | Number of parameters in the scenario | Use of parametric distribution | Use of solver parallelization | Solver | |
---|---|---|---|---|---|
Flux 3D | 0-75k | 0-10 | No | No | Direct (MUMPS) |
0-75k | 10+ | Yes, single machine or cluster | No | Direct (MUMPS) | |
75k-150k | 0-10 | No | Yes, see recommandations | Direct (MUMPS) | |
75k-150k | 10+ | Yes, single machine or cluster | Yes, see recommandations | Direct (MUMPS) | |
150k+ | 0-10 | No | Yes, see recommandations | Iterative (PETSc) | |
150k+ | 10+ | Yes, single machine or cluster | Yes, see recommandations | Iterative (PETSc) |