The same is happening for me.
Using Max 2021 - been using it for ages and recently was working on a very large Max scene that was almost unuseable with my Geforce 980Ti – 6GB card as a viewport card. Was running at about 5.9GB so everything was really slow.
Today I swapped out my viewport card for a 2080Ti which has 11GB vram so expected some 'spare' vram would be available.
NO - on loading the same scene, Max just gobbled up all the vram and is running at around 10.8 GB
So, - is it the case that Max will just take everything it can, load it into vram so it runs smoothly?
I have to admit that the viewport does run way smoother now, so that's good, but this does mean I can't render using Vray GPU or whatever.
I'm very hesitant to go and buy a 16GB viewport card as I imagine the same thing will happen.
I also tried unloading some of the XREF scene files that had ForestPack objects in – that helped, vram went down.
I saved the Max scene, re-loaded it now with fewer Xref scene files enabled, and gobble gobble, vram went all the way back up to 10.8 GB
AUTODESK - do you have an answer for this?
Using Max 2021.3.6, VRay 5.2, Forest Pack 7
Nvidia Driver ver 512.59 DCH