Yes, I know, it's an AMD forum and I'm posting several nVidia related things lately, but it is a very interesting design idea which runs contrary to AMD's approach, although necessary because nVidia and AMD are taking different routes to ray tracing, AMD with their programmable shaders, nVidia with dedicated silicon.
It certainly would explain how nVidia is supposed to achieve such greatly improved ray tracing performance on Ampere as it would happen on a separate chip, and would allow nVidia to reduce costs since they wouldn't have binning issues as great as they do now, plus it would explain the dual fan heatsink seen in the leaked images as it would be a dual sided PCB.
Also as the article states it would, at least in theory, allow nVidia to sell dedicated ray tracing cards, similar to the PhysX cards of old before their nVidia purchase and integration. Imagine the benefit for professional users, game developers, and television and movie studios, or even gamers who want a bit more ray tracing performance.
Divergent philosophies, who will come out the victor: Unified AMD, which is also perhaps about to tweak the Radeon branding slightly, or separatist nVidia...
https://wccftech.com/nvidia-traversal-coprocessor-rtx-3090-gpu/
