Dear Elmer users,
One of the most important topic in the recent development of Elmer has been the improvement of HPC capabilities of the code. New strategies for linear solvers have been implemented (FETI domain decomposition method, block preconditioners) and the parallel output has been improved with many reduction techniques and new output formats (parallel binary VTU format, HDF5 format). There is also an ongoing work to enable parallel mesh generation with Elmer. The work has been performed by several groups around Europe in a pan-European project called PRACE. Thus, it is fair to say that currently the public view of Elmer does show the true potential of Elmer for High Performance Computing (HPC). Simple problems with a billion unknowns have been solved and we hope that with the new strategies also more complex problems with tens and hundreds of millions unknowns become more of a routine.
Now to better reach the potential HPC users and to better understand the remaining bottle-necks, I encourage those stretching the limits of the current strategies to contact the developers. The aim in the size of the problem should rather be in 1000s of cores than 10s of them. You may for example, sent me a personal message describing your problem shortly.
Best regards,
Peter