Whats Your Reality Capture PC Setup?

Development Asset Creation
,
1

Hi All,

I just wanted to know what others out there are using as their PC to run RealityCapture. Any extreme builds or ideas about this using off the shelf parts? It might help with those of us upgrading to get a better understanding of what might be possible. I currently use Reality Capture for modeling laser scans of buildings & urban areas and already I’m seeing the limits of my own PC with “out of memory” alerts when working on large models. Also if the model is quite large I only see the end processed result as a point cloud in the 3D view due to limits of my PC.  Any ideas for better performance appreciated, thanks.

Below is my own setup:

CPU Processor - AMD Ryzen Threadripper 2990WX 32-Core Socket TR4 3.0GHz

Memory - Corsair Dominator Platinum RGB 64GB (4x 16GB) DDR4 3600MHz 

Graphics - EVGA GeForce RTX 2080 Ti XC GAMING Dual HDB 11GB Video Card

Storage - Samsung 970 EVO Plus 2TB NVMe 1.3 M.2 (2280) 3-Bit V-NAND SSD

Motherboard - MSI X399 Gaming Pro Carbon AC TR4 ATX 

2

Hi Anthony,
currently it is possible to display max 40M model in solid/sweet mode in 3D view (or less depending on your GPU Video memory). But you can use clipping box to display your mesh by parts.
You can find more information here:
https://support.capturingreality.com/hc/en-us/articles/360024840631-Warning-Not-enough-video-memory-

3

Anthony Milou replied to no-reply email so I am posting it back here.
Hi Zuzana,

Thanks for this. Increasing the RAM on my computer from 64GB to 128GB or higher will also allow a greater area of the site to be processed as currently with some large point clouds if I try to process all of it, it stops with a message about out of memory. I’m guessing its out of temporary memory used during the processing.

 

Kind Regards,

Anthony

 

4

Hello Anthony,

Increasing RAM will allow you to process bigger datasets, but won’t take any effect on displaying the result model in Solid or Sweet for this a GPU memory is used so you’re looking for an Nvidia Graphic card with a lot of GB. Rather than buying such card as Zuzana already mentioned I recommend you to use a clipping box to view your model in parts.

5

Hi Erik, Zuzana,

Thanks, with the below poincloud dataset I am currently able to process the area as marked in red using 64GB RAM, which is understandable given the high level of detail of the final processed mesh model (e.g see below fully processed reality capture renders at street level). If I wanted to process the whole pointcloud scene hopefully 128GB RAM will be enough to do the job.

 

 

 

6

Dear Anthony,
as I am guessing by processing you mean rendering it in “sweet” mode, then again this is just the cause of VRAM which is the memory of your GPU. If you lower your triangles count (simplify) you will be able to see the whole scene as in the pictures you have shown. You can just hit simplify in the reconstruction tab and reduce the polygon count to lets say 20 milion and then unwrap it and use texture reprojection which is again located in the reconstruction tab to bake your diffuse onto this new simplified model and you should be seeing both the whole model in solid and sweet and it should have the same color texture.