Quite interesting! Would be great to have these updates in the original source, lets hope.
This organisation has also very interesting videos online FYI: https://www.youtube.com/playlist?list=PLDPdSvR_5-GhR1OWQW3ovK_qGg09tmNuq
there’s another one from NEC who claims to be five times faster than SPOOLES and two times faster than PARDISO.
Interesting. The NEC link is described on the site: Applications?
I would guess that the NEC solution is using another linear solver which is proprietary software. On the other hand, the link from @rafal.brzegowy at the top shows a CalculiX code optimization. Is seems that I/O optimization has the biggest impact. If the code remains readable after such changes then such optimization is great. But Guido @dhondt would have to look at it and determine if it is doable/possible.
As described at NEC websites, they have two solver. Another one it has been posted on GitHub are public and opensource if I understand correctly. However I did not try to compile on Windows since no expertise in these field, may someone can confirm.
Did the guys from POP (see link from @rafal.brzegowy ) publish their source code improvements?
I doubt it… I looked for it when I saw the link but no mentions of source code anywhere.
More from POP:
Source code (some files):
@rafal.brzegowy do you know if they are posting the optimized source code somewhere?
I think it will be available in a new version (maybe from 2.18), because Guido Dhondt currently puts a lot of effort on CFD and the example here is also CFD. The report (link: https://gitlab.pop-coe.eu/documents/reports/-/raw/master/POP2-AR-021-CalculiX.pdf) clearly indicates who the applicant is.
This is just my personal opinion