I'm not sure this is really it, but matching an RTX2080Ti would be quite easy for AMD now, as demonstrated by the fact these results are more or less exactly what you'd expect from a 64CU, 384-bit Navi card.
An RTX2080Ti is 30-40% faster than a 5700XT on average.
If you made a 64CU version of the 5700XT that's 60% more cores.
If you gave it 16Gbps chips and a 384-bit bus width (Same as RTX2080Ti) that's 70% more theoretical bandwidth than 5700XT.
So even counting for practical losses and lower clock speeds a 50% gain over the 5700XT is pretty doable. Die size should be around an RTX2070 in that scenario, which is still a good chunk larger than a Radeon VII but not exactly out of this world.