this is one of the best technique i've found to actually have poin cloud direct on gpu from a 3d model. The system requires a good configuration of the 3d object , something like quadrimesh reflow in blender will help the conversion. The 3d element you are using need to have UV coordinate embedded : more accurate UV = better point cloud. This is where i've found those stuff : https://forum.derivative.ca/t/point-cloud-technique-missing-documentation/178985/7
This is what i ve done : https://www.youtube.com/watch?v=fCaIUIBpfRs. You can deform piece of o point cloud by putting a over layer to the unvwapped SOP and noise that little piece of texture. You can scale the resolution of the render, and by interpolation you get a lot of point froma low poly Model. Or you can adjust a photo Scan with the same technique by reducing the render size. Texture SOP can help sometimes to handel 3d object that does not have UV, but for some reason it breaks the object.Multiple object is REALLY EXPERIMETAL but works if you set it up correctly, 3 object is really hard to achieve. Have FUN!!!