There's that massive gap in performance when before it evolved more or less in tandem with the mean of transistor count and clocks (allowing for architectural changes). The 4070 Ti should have been 77% faster than the 3090 Ti by that metric but it's looking like a wash in compute and gaming.
I wonder if it could be a massive BEOL bottleneck or a huge dip in efficiency, which would be really silly... or maybe 5nm estimates are somewhat less meaningful (chip design is so automated, who knows, it might include redundant logic gates in post-design since the EUV won't pick up all of them, but that's included in the estimated count nevertheless).