I am currently working with large .vtu models in ParaView, and I am facing challenges with the visualization of increasingly larger datasets. The models typically range from 40-60 million grid points. The current setup, which includes a workstation with 16 threads, has improved visualization, but there are limitations, and I can only handle models up to approximately 3-4 GB.
To address these limitations, I am considering implementing a better GPU into the workstation. My question is whether upgrading the GPU would significantly enhance the visualization performance and allow for handling larger models, potentially up to 10 GB in size.
I understand that ParaView supports GPU acceleration through OpenGL and CUDA, and I would like to inquire about recommended GPU specifications for handling such large datasets effectively. Are there specific hardware considerations or configurations that you would suggest for optimal performance in dealing with models of this scale?
Any guidance or recommendations you can provide regarding GPU upgrades or other strategies to improve ParaView’s ability to handle larger datasets would be greatly appreciated.
Thank you for your time and assistance. I look forward to your insights.
Only through VTKm, and I dont think it is enabled in the binary release, which you will likely use.
Are there specific hardware considerations or configurations that you would suggest for optimal performance in dealing with models of this scale?
Some data representation (eg VolumeRendering) require your dataset to fit enterely in your GPU RAM, so you need to take that into account.
That being said, ParaView support distributed rendering, that let you split your data in multiple parts for processing and rendering, but it can be tricky to setup.
Any guidance or recommendations you can provide regarding GPU upgrades or other strategies
The fastest and bigger RAM, the best, especially if you want to use volume rendering or other GPU based representations (other than surface rendering).
I can only handle models up to approximately 3-4 GB.
I’m not sure what you mean here but you are probably not limited by your GPU unless you are using something else than surface representation.
I pondered your post, and think you are asking the wrong question. As mentioned by Mathieu, this depends on what you are doing, but I would be shocked if ParaView is compute bound of any form (algorithms or rendering).
As you don’t state what the issue is, I can only guess. You don’t even state what OS you are using. My guess is that either you are 1) load bound from disk or 2) running out of physical memory and starting to run into swap. To fix 1) get a big ssd, to fix 2) get a whole lot more memory and/or make sure you are swapping into an ssd.
Alan