Checking the numerical parameters for Plaxis calculations, one can choose as high as 256 cores. I am curious to know if Plaxis has ever been tested with so many cores and if Plaxis takes advantage of multi-threaded programming. I found on a paper (see ref blow), which dates back to 2014 that up to 8 cores were used and some limitations were observed for the Plaxis parallel linear solver when implemented for Toulon tunnel in France.
References:
Lingen, F.J., Bonnier, P.G., Brinkgreve, R.B.J., Van Gijzen, M.B. and Vuik, C., 2014. A parallel linear solver exploiting the physical properties of the underlying mechanical problem. Computational Geosciences, 18(6), pp.913-926.
Response for Plaxis support for Plaxis 2D 2017:
As far as I know, we never truely benchmarked our software above 8 cores for the overall process for the calculation process. The scaling when using more cores does not show a 50% reduction of calculation time from 4 to 8 cores. So I do not think that going from 8 to 16 cores. We have seen people use more cores, but I do not know exactly how this will scale. Also, the speed increase may depend on the type of calculation (plastic, consolidation, dynamics). When using more cores, the calculation kernel will need to divide the calculation over the different processors (cores/CPUs). However, this comes with some costs: the calculation needs to be split, and then the results should be merged.
Also, after some testing, we found out that a small number of elements divided over too many cores actually becomes slower to calculate. For that reason, we have a threshold value for the number of dofs (degree of freedom, dof) per core, which is about 10,000 dofs. So, if this number becomes too low, we just use fewer cores
not to slow down the calculation. So when using larger models, you will see an increase in the number of cores used.
For some more background on cores and performance, please see:
https://www.plaxis.com/support/tips-and-tricks/multicore-cpu-and-3264-bit/
https://www.plaxis.com/support/tips-and-tricks/is-plaxis-3d-enabled-for-multicore/ includes info about running parallel phases simultaneously. This might enhance the calculation speed dramatically.
As for the GUI operations, some parts actually scale well with more cores, but some other parts cannot be done in parallel, so the increase may very when using more cores.
Note that our calculation kernel was optimized for running on a single-core very efficiently before we started implementing parallelization some ten years ago. For true parallelization, it might mean we have to look at the base architecture to have better scaling with multiple cores (but that does not necessarily mean the calculation will become faster overall).
- Badges
- Science topic
- Similar topics
- Force