I think it's too late to start work on a voxel engine now. The future of 3D hardware is essentially an on-chip voxel engine allowing for real-time ray-tracing. Such hardware is already in development by Intel, Nvidia and AMD.
When such hardware is with us, groups of voxels will be fundamental graphics primatives and the way an engine that makes use of such hardware works, will be quite different from current approaches.
We can expect to see the dawn of such devices in around 2 to 4 years and with them we'll see the next generation of consoles making use of them too.
So spending a few years coming up with the fastest way to handle voxels on hardware poorly suited to the task will be made defunct as all the hard work is swallowed up by the hardware and the low level APIs.
True volumetric assets are much harder to create than traditional polygonal geometry - techniques that project and merge surface textures from inputted meshes to define the properties of internal voxels notwithstanding - there's animation and deformation to content with - which is the current showstopping bottleneck of voxel engines. Hardware will have some form of direct support for such operations.
The days of traditional 3D engines are numbered yes but not as much as the days for the current approaches to voxel engines.
It'll be the death of geo clip mapping due to tessellation hardware all over again.
Traditional engines will have slightly more shelf-life - running on older hardware, requiring far less time spent creating resources and potentially being faster on newer hardware for certain types of games (can't see entire planets made of voxels (with voxels smaller than pixels) being feasible for at least another 5 to 10 years)
Bookmarks