I’m working on a tile-based digital elevation model viewer/navigator of a nearby mountain range. Since it’s real world data, each tile is unique, so no instancing is possible. The maximum spatial resolution is considerable (0.5 m, which means 4 mio data points per square km / tile), so I’m using LODs, which still can exhaust the available physical memory of my machine. (Yes, I’m aware that LODs increase the amount of required memory.). For this reason I currently forgo the most detailed level and start with the second-most-detailed level, which cuts my memory footprint to one-fourth. Yet I’m regretting this when I’m zooming into one tile for a most detailed view.
So I’m wondering:
Is it somehow possible, to initially supply a low-res “dummy” mesh instead of the most detailed one to an LOD, and load the actual high-res mesh only on-demand?
At this point, any transient drop in framerate would be acceptable.