I think the new part here (and the reason it was at an NVidia conference) is that the computing power needed is now at a level to be able to do it real time, with a common GPU.