we are in almost the exact situation. Digital twin, AVP, Unity, all of it. Not industrial per se, but a need to render many many objects on screen. Would be interested if you had any updates on your testing, we are beginning testing this week. Thus far, spawning 4,000 default cubes or so cuts the frame rate significantly, and crashes shortly after. We'd like to explore combining the meshes to see if we can get a ton of tris on fewer objects. Maybe we will try shader magic.

To follow up on my above post, we managed to find a solution for data viz of more than 8,000 objects. A single mesh can contain millions of triangles before AVP starts to have a problem with it. For example, we added 1,000 cubes to a single mesh and duplicated this mesh object 1,000 times. That is 1,000,000 cubes, rendered at 90 FPS no problem. For our purposes, this is good enough. We can translate input coordinates to mesh-space and we can associate mesh-space coordinates with data outside of the objects. We can also color the verticies of the cubes individually, however we cannot put multiple materials on these mesh objects (unless each cube shares the same material in the object). We tried using sub-meshes in the object, and the results were much the same as if the meshes had been their own objects, hitting a limit of 8,000 sub meshes. We found the renderer struggled at around 2,000,000 cubes, or 24,000,000 tris. So, there is of course a graphical limit to what the AVP can render. Assuming the M5 chip is much better than this, we are testing on the original M2. The constraints are still present, but depending on the design these could be acceptible.