I'd argue that AMD messed up big time with the RDNA3 architecture, they told us 50-70% over a 6950XT yet in reality its just 35% so it looks like something went wrong with the move to chiplets.
They have also taken advantage of the ridiculous pricing by Nvidia to rename their card tiers so instead of a decently priced 7800XT we now have a £900 7900XT while the actual 7800XT we will end up getting looks like it will have just 60 CU's a drop from 72 on the 6800XT so anyone who bought an MBA 6800XT for just over 600 quid can forget about getting a performance upgrade for a similar spend this generation, the same for 3080FE buyers with whatever Nvidia decides to release at £650
Both of these companies are at it and are all to happy to fleece consumers, the GPU market is broken, Nvidia have a monopoly and AMD are just happy to toe the line rather than actually bring genuine competition.
MLID claims his AMD sources also say something went wrong. He says his sources don't work directly in on Radeon GPUs but that they work in other departments and get to talk with others in the company and in the months leading up to launch they were being told 100%, not 50%, 100%. That's why MLID came up with all the Nvidia killer rumours for RDNA3 was on expecting double 6950xt rasterisation performance.
But he doesn't know what went wrong. We can only make guesses ourselves but I'm pretty sure driver updates over the next 2 years ain't gonna add back 70% extra performance. If AMD genuinely believed they could be that much faster at some point, then some things went wrong:
* The 7900xtx was originally supposed to be the 7800xt and 7900xt is the 7800. So where is the real 7900xt? That would be the 128CU GPU that rumours said would be boosting to 3.5/4ghz
* so clearly something went wrong, a 128CU GPU never materialised and the clock speeds on the 96CU 7900xtx can drop as low as 1800mhz in games, no where near 3.5-4ghz
The only remaining question is why; why did AMD cut down the GPU to a fraction of its intended size and also why are the clock speeds so damn low, why are they even lower than rdna2
I don't know the answers, but what I do know is that the 7900xtx is power hungry; and based on that if such a 128CU 4ghz GPU was possible the TDP would be over 600w, so it's possible that once they started building samples they quickly realised they had a problem