The study has shown, that a significant speedup 
can be achieved when a finite element simulation is distributed over several computer cores. However, it depends on many parameters such as the discretisation of the problem, hardware configuration and employed software, to mention but a view. In order to study the different influencing parameters in detail, a 2d and a 3d boundary value problem have been modelled, which represent two typical usecases for the application of parallel computing. The 3d model has a large number of degree of freedoms (DOFs), while the 2d model requires very small increments in order to capture the high non-linearity of the problem. Both models require significant computation time when run on one core only.
 |
 |
| The mesh on top shows a cavity expansion modelled as a 3d problem with 177600 elements (647136 DOFs); the mesh on the right represents a plane-strain penetration problem of flat tip. This 2d mesh has been modelled with 25981 elements (183009 DOFs). |
The 3d model achieves a maximum speedup of =6.0, independent on the distribution of employed cores over the nodes.However, the average core load, i.e. the efficiency of a calculation, decreases with increasing number of employed cores.For the 2d model achieves a maximum speedup of =3.5, if all 8 cores of one node is employed for the simulation.
Previous observations from the use of parallel computing in project-related analysis can now be explained. Based on the findings of this study, recommendations for an efficient use of the available hardware can be given for different finite element models. Furthermore, more reliable purchase decisions can be made in the future.