It all depends on the size and number of voxels being used.
It is true that heightmapped poly-terrain has lower memory consumption but it is much harder of creating underground caves and such with it.
Also it is even harder creating of fully dynamically destructable terrain with heightmapped poly-terrain. Simply changing terrain height is easy but creating underground caverns, clifs, overhangs, and such is much harder.
It is true that voxel based models or terrain looks more like low poly models due to its cubigness but with some postprocessing you can improve that by replacing some of the cubes with other shapes which then alows you to get smoth edges.
One example of this is game Terraformer: http://www.vtss-llc.com/ It is true that the implementation of such postprocessing isn't best and that game itself has lots of other bugs, but you can see the idea of it.
EDIT: Another advantage of voxel based graphical engine is easy implementation of LOD (level of detail) as you can always join 4 voxels into one. And for doing this it is best to store data in octa-tree.
Bookmarks