You know what the problem is with RDNA 3? And i'm simply explaining what is going on, not excusing it.
The RX 7800 XT with 60 CU's, the same number of CU's as the RX 6800 is clock for clock 30 or more % faster.
The RX 7900 GRE with 80 CU's, the same number of CU's as the RX 6950 XT is clock for clock 0% faster.
The 7900 GRE with 18 GB/s memory is 0% faster than the 7800 XT with 19.5 GB/s memory, increasing the core clocks on the 7900 GRE does nothing to fix that, increasing the memory speed does, my 7800 XT hit's a hard scaling wall at 2900 Mhz, if i increase the memory to 21 GB/s, which is as far as it goes its scales again to 3100 Mhz, which is as far as i can push it.
RDNA 3 has a memory bandwidth issue, in that it needs it or it doesn't scale, at all, the only reason the 96 CU 7900 XTX is 45% faster than the 60 CU 7800 XT is because it has 50% more memory bandwidth, it has a 384Bit Bus, IMO with a 512Bit bus it would easily match a 4090, maybe it was meant to have 8 instead of 6 MC dies but they thought better or worse of it, depending on your point of view.