Netgen runs in parallel now

prop_design

I just noticed that the latest versions of Netgen run in parallel. You have built in options to control how many threads it uses. It definitely meshes a lot faster now. Also, if you leave the quailty plot open while it meshes, you can see the element quality improve during volume optimization. So that's really cool.

JohnM

I increasingly find myself relying on GMsh rather than Netgen. Very often the Netgen mesher gives me too much detail where a few splines have come together, and often does not produce a mesh when GMsh does. I do wish the refinement features that are coded in for Netgen worked for GMsh as well.

prop_design

I have the opposite problem. I can't get gmsh to give me detail where i want and it gives too much detail where i don't want.

prop_design

Update:

There's an article that they posted awhile back, that states you can expect about a 3x speedup.

https://ngsolve.org/news/new-features/52-netgen-parallelization

I didn't time it, but it definitely seems a lot faster, even with two cores.

prop_design

so on one model i tested i got a 20% speedup. this was comparing 1 core to 2 cores. all mesh settings were the same. my laptop doesn't keep a consistent speed. so any multi-core testing is impossible to predict or understand. but still, 20% is nice. it feels like it's much faster than that to me. the times are in the command prompt window of the attached pictures.

75.0915 sec for 2 cores
94.7519 sec for 1 core

the mesh has 315,639 nodes. far too big for me to solve.

this was using netgen 6.2.2003 available here; https://github.com/NGSolve/ngsolve/releases

you also have to install python 3.7 first. i used the latest version, 3.7.7, available here; https://www.python.org/downloads/windows/