Community
CFD Forum
Welcome to Autodesk’s CFD Forums. Share your knowledge, ask questions, and explore popular CFD topics.
cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Element count escalates quickly!

13 REPLIES 13
Reply
Message 1 of 14
esmaeilk
687 Views, 13 Replies

Element count escalates quickly!

Why does Sim CFD 2014 on my system counts 90 million elements for a simple model with 30 flat surfaces for a mesh size of 0.2? On other systems mesh sizes of 0.01 sum to about 1 million elements for much more complicated shapes. Is there something wrong?

13 REPLIES 13
Message 2 of 14
Jon.Wilde
in reply to: esmaeilk

Hi esmaeilk,

 

What happens when you actually mesh it? The mesh count preview can be innaccurate sometimes.

Once you have it meshed, the total mesh count will be shown in the Output Bar.

 

Kind regards,

Jon

Message 3 of 14
OmkarJ
in reply to: esmaeilk

Message 4 of 14
esmaeilk
in reply to: Jon.Wilde

I ran the simulation and it turned out to be about 3.7 million elements, however, I noticed that the system used 20GB of ram to generate the mesh volume. Isn't it supposed to be 2GB per 1 million elements?

Message 5 of 14
OmkarJ
in reply to: esmaeilk

~2GB/million generally applies for Solver, when the NS equations are being solved and iterations are run. Here, we are merely generating a mesh, and Mesher generally requires ~1GB/million.

 

At any rate, usage of 20GB for 3.7 million sounds extreme. Are you sure you didn't incorrectly measure the memory usage (extra 0 etc)? Also you should spot the memory usage when the message "optimizing volume mesh" is shown. If you wat until mesh is completed, the Solver kicks in, and the memory usage is escalated accordingly.

Message 6 of 14
esmaeilk
in reply to: OmkarJ

I'm pretty sure it was 20GB while creating the volume mesh and it took a few hours, I was staring at the screen all night! Is it possible that it has something to do with Multi-Threating?

My system has 12 cpu cores and 48GB of ram, I cant understand why the system cant run simulations ;(
Message 7 of 14
OmkarJ
in reply to: esmaeilk

Multi-threading (flag)  will help you in reducing the time required for the meshing rather than increasing it. Can you post an image of your domain? Also, it may be worthwhile to use coarse mesh initially and refine it gradually. You will have better grasp on memory usage and size of mesh. 

 

Also be advised that SimCFD uses only 2^n cores. Don't know if this will help, but try defining your Solver using 8 cores and se if this changes anything.

Message 8 of 14
esmaeilk
in reply to: OmkarJ

With a uniform mesh size of 2 for a small box you can see that all of the memory is exhausted and the system is now on page file/virtual memory.

screenshot.png

Message 9 of 14
OmkarJ
in reply to: esmaeilk

Well, you want to see the memory usage of only SimCFDMesher.exe, and not by the whole system!  See attached pic. Make sure you click on "Shw processes from all users."

 

 

mesh_memory.jpg

Message 10 of 14
esmaeilk
in reply to: OmkarJ

my bad, here it isscreenshot.png

Message 11 of 14
OmkarJ
in reply to: esmaeilk

In this case, I will be very surprised if:

a) You are able to finish the mesh in reasonable time (few minutes).

b) Mesh will have only 3.7ish million elements 

 

As you claim, a is false (as expected) and b is true, then this is strange and probably this question is for developers to answer.  

In your first post you mentioned 90 million elements in your earlier try. In latter trials the count was 3.7 million. So there appears an uncertainty in the trend.

 

I maintain that you start from a coarse mesh and go from there, so that at least you rule out the unreasonably fine mesh.

Message 12 of 14
esmaeilk
in reply to: OmkarJ

I have tried many different setups of meshing, even auto mesh.

Auto mesh number of elements are usually accurate but they don't result in smooth meshes even after multiple adaptations. And whatever manual mesh sizing that I do, the ram gets used far too much except for extremely coarse mesh. You can see that the CPUs are all idle.
Message 13 of 14
GubingWang
in reply to: esmaeilk

Hi, 

 

I encountered a similar problem and I read through this post, how was your issue solved finally? 

 

Thanks,

Gubing 

Message 14 of 14
OmkarJ
in reply to: GubingWang

During meshing, when the display message says "Optimising volume mesh...", observe the amount of RAM the process "SimCFDMesher" consumes. Approximately 1 GB RAM consumed corresponds to about 0.8 million elements. Thus, the RAM consumed can give you approximate idea about the number of elements and if you think it is unreasonably high, then just stop the meshing through Solve dialogue box and meshing will immediately stop. This methodology instead of waiting until the final mesh is mapped on the model will save you a lot of time.

 

As for having sufficient control on type (extrusion/structural), rifinement, quality and distribution of mesh, it remains an ongoing (but albeit very slow) process. I hope it improves in future. You may search in Idea Station and vote up the ideas you feel are important or create new ones if they already aren't there.

 

 

Can't find what you're looking for? Ask the community or share your knowledge.

Post to forums  

Autodesk Design & Make Report