Anyone here running Linux on dual core CPUs 2003

Unruh

THG reported no need for water cooling until ~4GHz. Customer reviews agree with this.

LS Dyna is not my code. It belongs to LSTC. The Appendix in their manual actually says at least 3 times that differences between the MPP and single processor results exist.

My own experiments using the exact same input files on a parallel supercomputing cluster and different single-processor desktop PCs confirm this.

AFAIK, no. In MPP finite element codes, the model is broken down into sub-components via domain decomposition. For each sub-component to be statically determinate, AIUI virtual supports must be added. The nature of these supports is especially relevant in problems with a high amount of model non-linearity.

Also, the time scale issue is important. Basically if a structure is deforming over seconds, you can linearise the problem over milliseconds or less. It may be that domain-decomposed models need a much smaller time step to linearise their system, compared to non-decomposed models on single processor machines. If this is true, then the parallel speedups may well be negated by the need for shorter timesteps.

The source code is copyrighted and not open source, unfortunately.

Thanks.