The title of this thread is wrong. If you actually read the article and watched the slides you would have noted that it isn't Sparse Voxel Octree GI, it is Sparse Voxel GI.
The presentation fragments in the article is far from complete, but the implementation they showed is a much simplified version compared to the technique invented by Crassin. They use Partially Resident Textures instead of Octrees, which removes tons of the benefits by the original SVOGI. No longer is the GI resolution variable, which means even more bleeding problems and no performance benefits when geometry consists of big open areas. They also have just shown a basic tech demo (where the low static resolution is apparent), so we haven't seen it running in a game, thus it might still not be possible from a performance standpoint in an actual game.
Also, since the concept is based on PRTs this will be fully feasible the next generation AMD and NVidia cards, which will support this technique.
