We don't need new CPUs; Reviewers still have to go out of their way with ridiculous, unrealistic settings, just to remove the GPU bottleneck that shows small differences between a budget i5 and a 12900KS. The overwhelming majority of games are GPU limited, Professional rendering is already moving to GPU-rendering, so the old days of Cinebench/Blender software render scores being relevant are numbered, and most productivity software is still bottlenecked by a single-core, if it's even multithreaded at all!
Between the 40-series and RDNA3, all the rumours are pointing towards Nvidia taking the monolithic, unbelievably expensive approach with giant dies, insane core counts, and ludicrous power consumption and MSRPs.
We know almost zero about RDNA3, other than they're going to be using chiplets - which have been a resounding success for CPUs and solve the scalability issues that have limited GPUs for well over a decade.
So yeah, my vote for RDNA3 is a no-brainer. If it's genuinely a working multi-die solution, it'll start a new era of GPU technology akin to the wake-up call that the CPU industry had in 2017 (and perhaps you'll remember how much the stagnation sucked before then with Intel refusing to move beyond quad core for EIGHT goddamned GENERATIONS!!)