Conclusion
Real-time 3D rendering is in a particularly exciting phase of its evolution. With the arrival of GPUs that are more and more programmable, a lot of approaches that have been out of the question up until now are reappearing, as developers try to solve the problems at hand.
Voxel octree ray casting is an especially promising technique for drastically increasing the geometric complexity of the games of the future. Obviously, it’s all still in the experimental stage, and Olick's demonstration at SIGGRAPH is more of a feasibility demonstration than anything else. A lot of questions remain unanswered.
One of those questions, which will be decisive for the future of the technology, has to do with the editing tools with which graphic artists will have to work. Olick says he expects an evolution of the system currently used for MegaTextures, which lets artists apply buffers to add details to a texture. Such a system, instead of modifying the normal map, would directly modify the geometry with this technology. But for now, everything is still in the conceptual stage and as we know, making nice demos is far from being the hardest part. They then have to be turned into a finished product that’s sufficiently stable and efficient.
Still, it’s very interesting to see a model animated using several million polygons (seven million in the initial model, done with ZBrush) at 60 frames per second on a present-generation card. These early results are extremely promising when you think about the hardware that will be available by the time games that actually use the technology hit the market. Given the development times for games nowadays, and considering how id Software doesn’t plan to release Rage before next year, we’d probably have to wait three or four years before this type of rendering hits retail shelves.
