Laggy point cloud manipulation after LOD acceleration structure operation

[foresterwill]

Hi,

I've been having issues with manipulating a large (~28GB) point cloud in the program. The cloud has around 900 million points. Initially, after the cloud is loaded into the program, there will be no issues with the framerate when rotating, panning, zooming etc. However, it appears that after the LOD process has finished, the cloud becomes very laggy and difficult to work with. I have ticked the option to decimate the cloud when moving it, and enabled GPU cloud loading, but this issue persists. Around 14GB of the video card memory and 50GB of the system memory is being used. I have an RTX 3090, 128GB of RAM, and a 16 core processor, so I find it implausible that this is a hardware limitation. Let me know your thoughts.

[daniel]

Hm, that's maybe because the LoD strategy is counter productive on your setup?

Can you simply deactivate it? (so that it works as when the cloud is freshly loaded)

[foresterwill]

Thanks for the reply Daniel,

I have tried to disable the LOD feature, but I can't find an option to do so. Is it somewhere in one of the drop down menus?

[daniel]

Go to 'Display > Display settings'.

Then uncheck the 'Decimate clouds over XXX points when moved' option.

[foresterwill]

Doing that actually made it worse. My GPU cannot handle all 900 million points at once. When I check the downsampling option (~150 million), I can initially move the cloud around, but then the LOD starts computing levels, and by the time it's finished, the lag makes the cloud pretty much unuseable. Any ideas here?

[daniel]

Not sure to understand: if you disable the 'LoD' option (uncheck the checkbox in front of 'When moved, ...'), then you should stay in the initial state you were in (when you said 'after the cloud is loaded into the program, there will be no issues with the framerate when rotating, panning, zooming etc. ').

And the LoD structure won't be computed. Technically, there's no need to fit the whole cloud in VRAM (CC will store what it can, and it will keep the rest in standard RAM).

[huhug]

Hi Daniel,

I have exactly the same problem. When manipulate a large point cloud (a 10 GB E57 RGB cloud), after the acceleration structure is calculated, everything just become super laggy and the program doesn't seems to call GPU anymore (observing from the task manager). I tried your suggestion for unticking the 'Decimate clouds over 50.0 M. points when moved' option and reloaded the point cloud, it just got worse, and then I ticked this box again on the fly, everything becomes smooth again, until the acceleration structure is calculated.

So to summarize, the point cloud can only be smoothly viewed when (1) 'Decimate clouds xxx' option is ticked (2) the acceleration structure have not yet been calculated.

Appreciate a lot for your help!

[daniel]

That's weird...

2 questions:
- what graphic card do you have?
- what about the 'try to load cloud(s) on the GPU...' option? Is it checked? Does it change something if you change its state?

[huhug]

Dear Daniel,

To your questions:
1. I have a Nvidia RTX4090
2. The option 'try to load cloud(s) on the GPU...' is checked. I tried to uncheck it and it becomes super laggy again.

What I have noticed newly:
When the acceleration structure is not yet calculated, although the visualization (moving around) is quite smooth, when I tried to zoom into details, there are limited points shown. After the acceleration structure is calculated, the visualization become very laggy, but zooming into details, there are much more points shown. It seems that it was only very smooth in the beginning because only part of the point cloud is shown?

What do you think?

Cheers

[daniel]

Frankly I don't know... nothing really makes sense. I guess you are using the latest driver?