@Jim I asked him about chiplets design for GPU, what his take was on it. He said that having multiple "gpu" chiplets would just make it like a crossfire solution, which would not be so beneficial as we know. Also if you would go to towards a good working solution that would work on a chiplet design (I assumed like breaking up the inner parts of a GPU similarly like Zen 2 with its I/O die and compute dies) it would be very very... very difficult he said.
My follow up question was what if you had one full Raster GPU chiplet and several pure Raytrace chiplets (because ray tracing by nature is highly scalable, near perfect), if that would that be beneficial. His answer started out positive but we got soon interrupted by the stage before he had the chance to share this take on it 😀
I didn't went back to him unfortunately to continue that discussion and I also had other similar interesting GPU questions, kinda regret it now ....but had to talk to others at the party as well so 😉