You can actually see how that rumour started though.
So someone gets some info that N33 is 20+ Tflops which is about what the 6900XT is, leaks the info, someone assumes that means N33 must have double the shaders over N23.
On top of that you get the info about RDNA being architected for 3Ghz + so you think, 4k shaders + 3ghz or above clockspeeds and suddenly the idea that it could match a 6900XT at low resolution where the 32MB and 128 bit bus are not such an issue begins to form.
The reality is though it does not have double the shaders, they are just dual issue and it seems they barely work in games but give it a compute workload and you can get 21Tflops out of it.
Also AMD have form for this kind of thing. The 4870 had 2.5x the shaders of the 3870. RDNA2 was a huge leap over RDNA1. Zen3 was a big uplift over Zen2 on the same node so given their recent track record of exceeding expectations people thought RDNA3 would be more of the same. Obviously it didn't turn out that way but it was a fun ride while it lasted.
This shows how stupid people are and lack critical thinking. We knew Navi 33 die size and manufacturing node for a while now.
How would AMD raise or maintain IPC, while shrinking the die on what is largely the same manufacturing process, while doubling the shader count( and tflop count)? Particularly without the xfactor chiplet presents(at the time). Particularly with barriers which should limit performance. People somehow expected the impossible possible.
Some of these impossibilities included violating the laws of physics(double transistor density while not running into heat issues on the same node), fix some glaring flaw which made RDNA2 inefficient in regards to performance per die space(sarcasm). Magically get over bandwidth barriers with only l3 cache only 1 quarter of the size.
No one thought of these potential and obvious roadblocks which should temper their expectations(atleast in the AMD fanbase). It should have been obvious that AMD was doing something very cheap in terms of silicon which meant IPC was going way down(gaming performance per teraflop). Being able to to more than double performance, while not increasing die space, on the same node with some severe bottlenecks in place is impossible, particularly with how well RDNA2 was designed.
But people ignored this and all I can blame is the tribalism effect. People just following each other, believing a youtuber, while not evaluating if it's true or not. Simultaneously not listening to reason which is valid because they would rather feel good and be stupid.