You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
From what I can see with Industrial simulations that I have run and others have run (see forums https://forum.cadet-web.de/t/cadet-match-sma-simulation-not-converging/364/11) solving in parallel is almost always much slower. In the case of the forum the simulations ran about 10x faster by setting the mode to 2 and most people don't know about these settings and end up with very slow running simulations.
The text was updated successfully, but these errors were encountered:
Do you recommend defaulting to linear mode completely, or do you recommend raising the barrier?
In case of the latter, what is your suggestion for the switching point?
This is a good question. So far, I have not run into a single model where parallel solve was better when there were no cycles. I will ask Johannes if he has an example that runs better in parallel with no cycles. I suspect if you have multiple 2D GRM in a chain the linear solve would still be faster because you wouldn't have to repeat solve such expensive unit operations, but I have not tested that since I don't use 2D GRM.
I'm bumping the barrier up to 25. That is, sequential solution mode is selected for systems with up to (including) 25 unit operations. Please let me know if this is still insufficient.
From what I can see with Industrial simulations that I have run and others have run (see forums https://forum.cadet-web.de/t/cadet-match-sma-simulation-not-converging/364/11) solving in parallel is almost always much slower. In the case of the forum the simulations ran about 10x faster by setting the mode to 2 and most people don't know about these settings and end up with very slow running simulations.
The text was updated successfully, but these errors were encountered: