topic Re: Solve time in CFD Forum

Solve time

Anonymous — Sat, 11 Apr 2015 16:46:08 GMT

Hi all,

After refining the mesh for a model I have just set up there is a total of around 4 million nodes existing. I have read online that the required RAM usage per million nodes is around 1 - 2.5 GB.

Can anyone give me an estimate of how long it is likely to take to solve this senario on a computer with 8GB of RAM?

I have set the solver off but it has been generating the volume mesh for around 1.5 hours now and there is no progress-bar to guage where the computation is up to. I am a little worried it has crashed or something so just thought I would ask.

Any help is much appreciated

Thankyou

Re: Solve time

Jon.Wilde — Mon, 13 Apr 2015 07:34:11 GMT

Hi,

It sounds like it is still trying to mesh but might never get there. How much RAM is it using?

We can't really estimate how long a job will take to run, this is dependend on the CPU and the model type.

Most fairly complex models only take 10-30 mins to mesh though.

Have you checked your mesh diagnostics?

Kind regards,

Jon

Re: Solve time

Anonymous — Mon, 13 Apr 2015 10:07:04 GMT

Hi,

It was using the full 8GB of RAM available on my computer. It meshed in the end and stated solving but I had to stop it at aroud 40 iteration becuase it was only running through around three iteration per hour.

Is there anyway you can access the cloud service to run the simulation from there as a student?

Cheers

Anthony

Re: Solve time

Jon.Wilde — Mon, 13 Apr 2015 10:08:40 GMT

Hi,

If it was using all 8, it is likley to have also been using a page file, which can be incredibly slow.

Do you need that much mesh?

Feel free to share a CFZ (support file) and I can take a quick peek.

As far as I know, the cloud would need to be paid for to gain access.

Thanks,

Jon

Re: Solve time

Anonymous — Mon, 13 Apr 2015 10:33:25 GMT

Hi,

I tried a model with only around 400,000 elements but it didn't really give me the results I wanted. I'm trying to analyse the jet wake profile between the impeller blades, and therefore I used a 0.5 refined mesh on the blade suction surfaces and leading edges, used a uniform mesh in theses regions and then spread the changes. I also used a uniform mesh on the inlet and outlet of the RR. This resulted in around 4 million elements.

On the first model - 400,000 - the secondary flows were not captured. Also, the impeller didnt seem to push the fluid around the volute as expected. I tried to checked why this was with the trace function and found that the traces didnt go through the RR at all; stoping at the end of the inlet or if added at the outlet stoping at the RR. Is this usual?

I have removed both jobs now so unfortunately cannt provide the support files but this is good to know.

Below is a pic of the simple assembly so you have an idea what i am testing. I have also provided a cut plane for reference of the internal geometry. I have tried to make it as cfd ready as possible - no small fillets, unecessary edges etc. The fluid is compressible

Cheers John

Anthony

Re: Solve time

Jon.Wilde — Mon, 13 Apr 2015 10:58:06 GMT

Ah yes, apologies - jugging so many models 🙂

These do tend to need a good level of mesh and processing power, there is no avoiding it really.

I would look at results on a cut-plane, just to be sure of what is occurring. Traces can be impacted by the mesh.

Thanks!

Jon

Re: Solve time

Anonymous — Mon, 13 Apr 2015 12:26:24 GMT

ok thanks John

Re: Solve time

Anonymous — Mon, 13 Apr 2015 21:20:29 GMT

Hi John,

Sorry to bother you again but would it be possible to have a quick peek at this support file as you suggested just to see if it is using a page file? Every time I try and solve anything it is impossibly slow and my computer doesnt have bad spec.

King Regards

Re: Solve time

Jon.Wilde — Tue, 14 Apr 2015 13:59:14 GMT

Hi,

I can't do this as our machines are different and mine might not need to use a page file. You can check it on yours though - take a look in the task manager while it's meshing/running and at the bottom you should see whether the page file useage jumps up. This is all I can think of at the moment,.

I have taken a look at the model though and definitely have some suggestions.

Your mesh is incredibly fine, you could mesh this coarser for sure. Perhaps a manual mesh with a finer mesh just on the leading/trailing edges might be more sensible.

Pop the surfaces into groups to help with quick selection. These surfaces can then be given a finer mesh than the Rotating Region.

My manual mesh looks like this, 1.8m elements and should be a good starting point. You could run a mesh sensitivity study from here.

Re: Solve time

Anonymous — Tue, 14 Apr 2015 19:01:31 GMT

Thanks ever so much for your support as always John, its much appreciated. Will give this simulation a try with your modifications and see it goes

Thanks again

Anthony