Learning Forum

[v.bhatnagar.1]

Hello all,

I have been running a viscoelastic extrusion simulation (evolution of relaxation time) on Ansys Polyflow on my University's HPC. It runs properly, but very slowly - 3 days and 20 hours for 4 steps so far. I am currently running it on 8 cores. As I plan to increase the number of simulations and also the domain size, are there practical steps that I can use to speed these simulations? Also, are there ways to reduce RAM usage?

Looking forward to your help! 

[Prashanth]

Hi Vedanth,

Viscoelastic simulations do take longer to run, there's many more equations than any a simple Generalized Newtonian case. If you reduce the grid size by placing the cells judiciusly, you can speed up by not giving up on accuracy much. Also, if you have a GPU card that is supported by the GPU accelerator feature, you can use it to boost the computatioal speed. Another option is to do a distributed memory parallel run on MUMPS solver and run the simulation on multiple nodes.

[v.bhatnagar.1]

 

Hello Prashanth,

Thanks for your reply. I am using Ansys 2024 R1. I will try to reduce the number of mesh elements.

Some queries:

1. Is it possible to tweak the solver (numerical or evolution parameter) settings to accelerate the solution?
2. How can I use the MUMPS solver?
3. How does GPU acceleration work? Does the GPU VRAM need to be as large as the mesh size? Can GPU acceleration work with regular RAM and alongside the CPU? Is there a published comparison between running Polyflow (or Fluent) on a CPU vs. GPU? Please shed some light on this.
4. Can you please recommend a GPU / CPU combination(s) that is known to be compatible with Ansys Polyflow (Ansys 2024 R1)? I can look into procuring a workstation with your recommended configuration. Our budget would be a maximum of 3000 GBP.

I will be performing about 200-300 viscoelastic co-extrusion simulations. If you can answer with this perspective, that would be great.

Please let me know if you need any information from my end.

[Prashanth]

Hi Vedanth,

Here are the answers to your questions in the same order you wrote them, applies for both Fluent Polyflow and Polyflow Classic..

1. It differs for each and every case. To save some calculation time - If you are running a steady evolution simulation, and if the calculation runs to completion without any failed in-between steps which caused the evolution parameter to decrease, then you can experiment with larger evolution step sizes, larger initial step size etc.. There are some evolution recommendations already listed in the Polyflow user literature: 27.4. Using Evolution
2. The MUMPS solver (a recent addition to Polyflow) is used to run very large cases (grid size in millions) which needs to be run on multiple nodes using distributed memory parallel (DMP) analysis on cluster. When you don't run a distributed memory analysis on multiple nodes, even when using shared memory, it uses twice the memory as the default AMF solver. See this user literature section on how to activate it: 30.2.7.1. Selecting the MUMPS Solver
3. The GPU acceleration has been out for a while - note that it is not that the entire Polyflow's calcualtion makes use of the GPU, only some part of the calcuation. It merely boosts the calculation speed by a some percentage when we use a supported GPU along with the usual CPU for workstation runs. We will not recommend it - if you can avail, a cluster with many CPU nodes for a large viscoelastic case run is more suitable when you need to run hundreds of large simulations - the GPU accelerator only supports one CPU at a time, so only for local workstations - it is way fatser to use a parallel DMP analysis on multiple CPU nodes on cluster. Must read sections: 30.3. Distribute-Memory Parallel (DMP) Analysis for Ansys Polyflow Classic ; 30.4. DMP Analysis on a Cluster. And then, all the information you need to know about the GPU accelerator is here: 1.7. GPU Accelerator capabiity.
4. For many different reasons, I cannot suggest you a workstation confirguration within that price bracket. But I can give you the information that will help you make that decision. Platform support > GPU Accelerator capabilities 2025R2 > Polyflow.                                                                                                                                                                                                      Also, this GPU accelerator capability as of now is supported by Nvidia cards. We have some recent tests using RTX6000, and we have also seen others using RTX4000 for their workstations. So, there shouldn't be any problem with similar series cards. Also, if you are planning to use GPU accelerator feature in your windows workstation, it is strongly recommended to upgrade to atleast v24R2 and above.

[v.bhatnagar.1]

Thank you very much. These comments are greatly helpful.