when rendering voxels in octrees, the only information necessary is the current octree-level, -position and colour/texture. But one has to send eight vertices to the rendering pipeline in order to render a complete cube and still needs colour/texture.
Is there a better way?
Since geometry-shaders wont be wide-spread for quite some time the "emit_vertex"-method in the geometry-pass wont suffice (or am i wrong? i don't think mesa currently supports geometry-shaders). Also, afaik, backface-culling is done before the geometry-pass, so that would lead to additional overhead.
My current idea is to send 'dummy vertices' and have the actual drawing data in between but that really shouldn't be the solution.