[Echidna]

I run Ansys CFX with an Intel Core i7 920 (4-core, 8 threads).
I would like to ask why the processor is never used to its maximum power. Whenever i make runs in CFX, i get only up to 13% load on the processor. Is there any way to exploit the full power of my CPU, so it will run close to 100% load? I am attaching a print screen picture to let you see.

[Lance]

Maybe a silly question but are you using Run Mode: Serial or Local Parallel in the solver ?

[Echidna]

Hi Lance,
I used the defaults, can you help me on these options?

[Lance]

Um, that explains it I guess.
Try PVM Local Parallel or HP MPI Local Parallel (see documentation!) and set the number of partitions to 4 and see if you use more cores. I think there is a tutorial when local parallel is used, have a look at that.

[Echidna]

Thanks Lance, i used partitioning with both HP MPI local parallel and MPICH2 Local parallel and i got an error. See the attached picture.

[Lance]

Looks strange. Dont really know what to do if the simulation runs in serial but not in parallel. The error is something about vertex, mesh problem?

[Echidna]

That's the problem reported. The exact same analysis runs OK in serial.

+--------------------------------------------------------------------+
| Partitioning Information |
+--------------------------------------------------------------------+

+--------------------------------------------------------------------+
| ERROR #001100279 has occurred in subroutine ErrAction. |
| Message: |
| Signal caught: Segmentation violation |
| |
| |
| |
| |
| |
+--------------------------------------------------------------------+

+--------------------------------------------------------------------+
| ERROR #001100279 has occurred in subroutine ErrAction. |
| Message: |
| Stopped in routine FPX: SIG_HANDLER |
| |
| |
| |
| |
| |
+--------------------------------------------------------------------+

+--------------------------------------------------------------------+
| An error has occurred in cfx5solve: |
| |
| The ANSYS CFX partitioner exited with return code 1. |
+--------------------------------------------------------------------+


This run of the ANSYS CFX Solver has finished.

[Echidna]

Anyone please?

[Lance]

You'll have to provide more information about your case. What are you modeling, types of mesh, etc.

And also, can you run the tutorials in parallel?

[Echidna]

The results refer to "Overflow". Maybe this helps a bit?


+--------------------------------------------------------------------+
| ERROR #001100279 has occurred in subroutine ErrAction. |
| Message: |
| Floating point exception: Overflow |
| |
| |
| |
| |
| |
+--------------------------------------------------------------------+

+--------------------------------------------------------------------+
| ERROR #001100279 has occurred in subroutine ErrAction. |
| Message: |
| Stopped in routine FPX: C_FPX_HANDLER |
| |
| |
| |
| |
| |
+--------------------------------------------------------------------+

+--------------------------------------------------------------------+
| An error has occurred in cfx5solve: |
| |
| The ANSYS CFX solver exited with return code 1. No results file |
| has been created. |
+--------------------------------------------------------------------+

End of solution stage.

[Echidna]

Originally Posted by Lance

You'll have to provide more information about your case. What are you modeling, types of mesh, etc.

And also, can you run the tutorials in parallel?

Lance, thanks for your time.
I make an analysis for a simple racecar diffuser.
Please look at the attached pictures.

[ghorrocks]

The 13% CPU load is simply because you are running single processor. You have 8 cores (well, probably 4 with 4 virtual ones), so a single core is 12.5%. Run parallel to access more CPU load.

You have a partition problem, but it look like you got past that yourself.

A floating point error just means the simulation diverged. You need to improve numerical stability - better mesh quality, smaller timesteps, double precision numerics, better initial conditions.

Also, why are you not taking advantage of the symmetry plane of the geometry? Also the very fine mesh along one edge of your geometry suggests you meshing is poor. Did you intend to concentrate the mesh there?

By the way, there is nothing simple about a race car diffuser. You will have to know what you are doing to get this analysis accurate.

[Echidna]

TBH, i am doing my first steps in the world of CFD, the mesh concentration at the front edge of the diffuser was not intended. Most of the meshing options were left as default. I am currently working on learning efficient meshing in CFX.
And yes, a race car diffuser is a delicate and important "instrument". By writing "a simple diffuser" i meant that my design is simple, it wasn't an advanced design of a diffuser.

[ghorrocks]

I would fix that meshing problem on the edge before proceeding. That is going to double the size of your mesh and cause lots of problems.

While you are at it, cut the geometry in half and use symmetry. That will help a lot too.

I am glad you realise a diffuser is a delicate "instrument", my point was more that an accurate CFD analysis of something like this is also a challenge. Don't expect to get good results quickly if you are new to CFD.

[Echidna]

A last question: While i can overcome the problem of CPU usage in solution by using parallel solution, while meshing the CPU load is again 13% maximum. Any way to overcome this and exploit all the CPU power during meshing too?

[ghorrocks]

Exactly the same issue - meshing runs on a single core by default. There is a parallel meshing thing available but I generally don't use it. If you are generating meshes which require parallel meshing then most of the time something is wrong. Fix the problem (see my previous post for what is wrong) and it will mesh quickly on a single core.

[Echidna]

Thank you Glenn!