Among other things, they discussed three key parts. First, Neural Arrays tech lets GPU compute units work together as one AI engine, which helps things scale better and allows bigger machine learning models to run with less overhead. Second, Radiance Cores add special hardware for real-time ray tracing and path tracing, which makes light transport better and frees up the GPU to focus on shading leading to faster and more efficient rendering. Last but not least, Universal Compression squeezes all data moving to memory—not just textures—which cuts down on bandwidth needs while giving higher detail better frame rates, and improved power efficiency. However, all of those are just on paper or in software simulation mode so we will have to wait a few more years until they become a reality.
