According to a video by tipster Coreteks, NVIDIA's upcoming Ampere GeForce RTX 3080 and related cards could be set to feature a traversal coprocessor, separate from the GPU die, to massively accelerate hardware ray-tracing performance. NVIDIA will purportedly use a dual-sided PCB design to implement the coprocessor: it will sit opposite the GPU core.
This approach has significant advantages compared to NVIDIA's current strategy of stuffing RT cores alongside standard CUDA cores on-die: dedicating the entire GPU chip to raster workloads gives NVIDIA an easy way to deliver substantial gains to Ampere graphics performance, even without major architectural changes relative to Turing.
NVIDIA has been playing with the idea of a traversal coprocessor for quite some time, having submitted a patent application for the technology as far back as 2018.The traversal coprocessor could be how NVIDIA manages to improve ray-tracing performance to the extent promised by the rumour mill: RTX 2080 Ti levels of ray tracing in the mid range RTX 3060.
We've heard very little so far from NVIDIA regarding its consumer-oriented Ampere Geforce graphics cards. However, with the impending arrival of the PlayStation 5 and Xbox Series X, as well as the launch of AMD's RDNA2 Big Navi lineup, we should hear more soon.