[Dennis Pongi]

HI,
I am an aerospace engineering student.
I need a new PC, I decided to try to build it myself. Using it for CAD design and CFD simulation. I have to choose which components to direct my budget towards (max 1500 euros). (The university does not allow the use of the server)

The software I use is Star-ccm+, with the latest version it also integrates GPU simulations for Windows systems.
-Assembly with 200 parts
-Simulations with 20 million cells

After reading many articles I ask myself this, does it make sense to invest in a good GPU to integrate GPGPU calculation (even in the future)?
Is it possible to integrate video cards like NVIDIA RTX 4070super with combined CPU-GPU calculation?

If it is possible, does it speed up the resolution time?

If that's not possible, which graphics card would you recommend? Is it worth evaluating an Nvidia RTX 4060?

For the other components I already have a pretty clear idea, simplifying:
RAM 64GB
CPU: >16cores
Abundant cooling

Thank you

[wkernkamp]

Originally Posted by Dennis Pongi

HI,
I am an aerospace engineering student.
I need a new PC, I decided to try to build it myself. Using it for CAD design and CFD simulation. I have to choose which components to direct my budget towards (max 1500 euros). (The university does not allow the use of the server)

The software I use is Star-ccm+, with the latest version it also integrates GPU simulations for Windows systems.
-Assembly with 200 parts
-Simulations with 20 million cells

After reading many articles I ask myself this, does it make sense to invest in a good GPU to integrate GPGPU calculation (even in the future)?
Is it possible to integrate video cards like NVIDIA RTX 4070super with combined CPU-GPU calculation?

If it is possible, does it speed up the resolution time?

If that's not possible, which graphics card would you recommend? Is it worth evaluating an Nvidia RTX 4060?

For the other components I already have a pretty clear idea, simplifying:
RAM 64GB
CPU: >16cores
Abundant cooling

Thank you

The size of your problem is quite large. It may fit within 64GB, but I would be more confident with 128GB. Note that the GPU also needs to access the data, so it matters how much GPU memory you have. The GPU speedup will diminish if there is a lot of traffic between GPU and CPU memory. If it were me, I would focus on a CPU based calculation setup with enough RAM and highest possible bandwidth. You should look at openbenchmarking.org openfoam large problem solution times to have an idea what are good setups.

[Dennis Pongi]

Thanks so much for the advice