Learning Forum

[andrzej.tusk]

Hello 

I'd like to upgrade Graphics card in my Ansys Discovery Live workstation.

Currently I have Quadro P5000.

 

Calculations often have many low thickness walls with internal fluid flow.

 

I'm thinking GP 100. I have budget of around 5000 - 7000 USD

 

What would be best choice efficiency vs cost

[Naresh Patre]

Hello Andrzej Tusk

Quadro P5000 and GP100 both seem to have 16 GB GPU memory while the latter one has higher CUDA cores. If the thin regions in your model are not getting resolved with Quadro P5000 model, I doubt they will get resolved with GP100 as Discovery Live has some limitations while handling thin regions. Please check out below post on handling thin geometric features in Discovery Live If you haven't done so. Meanwhile, I will also check with the team and will let you know.

Discovery Live: "Thin" geometric features and relative GPU adjustments - Knowledge Base - Ansys Discovery Forum

[Naresh Patre]

Hello Andrzej Tusk

I got the same response from team that it cannot be guaranteed whether upgrading to a higher GPU memory (from Quadro P5000 16 GB GPU memory) will capture thin regions in your model. As mentioned in the "thin geometric features" post shared earlier, please check if you are able to reduce the domain size for capturing thin regions.

[andrzej.tusk]

Hello Naresh Patre

Thank you for your response.

So GP100 is not the way. What about two nv Linked RTX cards or GV100 or P6000 will this help?

[Naresh Patre]

Hello Andrzej Tusk

It's difficult to tell if any of the higher GPU memory cards like GV100 (32 GB) will be able to capture all thin sections in your models. Currently Discovery Live has some limitations while handling thin regions.

[andrzej.tusk]

Hello Naresh Patre

At the moment Discovery Live captures my thin regions well, but fidelity settings has to be high (top 75%+). I'd like to achieve one of two effects with this upgrade:

1) Faster calculation with current fidelity setting

2) Ability to lower fidelity setting while still capturing thin regions therefore making calculation faster

 

At the end higher calculation speed is the target.

 

Can any hardware upgrade support that?

[Naresh Patre]

Hello Andrzej Tusk

Good to know that the thin regions are being captured. How much time (elapsed time on your machine and not the simulation end time) does it currently take to get the results with highest fidelity?

[andrzej.tusk]

Hello for some models we wait up to 2 hours to see indication of the results. If we want really fine results we go for 8h. We mostly build air handling devices made of sheet metal. 

[Naresh Patre]

Hello Andrzej Tusk

Thanks for sharing the details. I am checking with the team and will get back to you. Meanwhile, Do you monitor any solution parameters (like outlet velocity) through charts? Does the graph stabilize after 2hrs / 8hrs, or do you see the values stabilized much earlier? If the latter is the case, you may consider the results as finalized and may want to stop the solution from solving further.

[Naresh Patre]

Hello Andrzej Tusk

Is it possible for you to share your model for the development team to look into? You can attach it here which could also be visible to all forum users. If the model is confidential, I can contact you separately for sharing the model. Let me know.