I saw many voxel game developers opt for a strategy that involves using an octree whose leaf nodes include a 16x16x16 (or 32^3 or 64^3) chunk instead of having leafs represent a single voxel.
What is the reason for this? I also don't understand how this will coincide with LOD, if the leaf is a 32x32x32 chunk, how does the tree manage LOD levels inside the chunk?
If it doesn't, then wouldn't you lose a lot of detail when considering this chunk as 1 voxel? I can't wrap my head around the benefits of doing this :)
I implemented my own version of Octree (Single voxel leaf) with face culling between nodes and I can get the following result (512 ^ 3 voxels):
This takes about 50 MB of RAM (60 FPS) and when transitioning to 1024 ^ 3 it jumps up to 190 MB of RAM (60 FPS).
I am trying now to optimize memory that's why I wanted to start experimenting with LOD but the more I researched the more I found mentions of chunks as leafs.