Today, material non-linearities, large strains and deformations, high-velocity impact problems are almost routinely treated by sophisticated finite element tools.
The progress in the field would be unthinkable without the availability of computer platforms, especially the parallel ones.
However, parallelism shows a new worrying phenomenon appearing with the implementation of finite element codes on MPP machines: The dependency of results on the number of CPUs used.

The figure represents results of a car crash simulation – a comparison of engine bottom accelerations computed using different numbers of CPUs and a commercial software (taken from a Proper Misuse of Finite Element Analysis by Miloslav Okrouhlik).

We should not believe in a dependence of FE results on the number of CPU’s used and we should admit that a good parallelization of nonlinear dynamic problems can be difficult to master.