How waas Vega 7, that was running on 7nm process, but could not even dream of 2080Ti levels of performance, on top of being 1.5 times behind on perf/watt front, "vaguely competitive"?
Only architectural changes, RDNA1&2 allowed AMD to first close the gap and then beat NV.
Is "within 10%" not vaguely competitive to you? Vega64 went from being 30% behind a 2080 to a Radeon VII being just 9% behind a 2080 through the move to TSMC:
You're still also comparing a 2017 architecture designed for GloFo 14nm (not TSMC 7nm) to a 2018 architecture designed for the outset for the TSMC 12nm mode it was mode on. If you've followed any CPU/GPU design in the last decade or so, you should know well that porting a design from one process node to another is not the best way to make a chip and gets far less from the new node than a new design made for that node specifically. AMD, Intel, and Nvidia have all experienced die shrinks with less than ideal gains over the years, which is why the rather significant improvements between Vega64 and Radeon VII were so noteworthy.
I'm not really sure why you're even focusing on the Radeon VII, it was a low-volume part that didn't have any product stack underneath it, lacked many of the features that Nvidia was offering, and simply served to provide
something as a placeholder whilst RDNA was being worked on as the designed-for-7nm part, not a 2-year-old port from a different process node and foundry altogether. GPUs are also only half the equation here - TSMC 7nm is also responsible for giving AMD an advantage over Intel in the CPU department, With Zen2 offering significant leaps over Intel at the time, and being a massive step up from the Zen/Zen+ on GloFo 14 again.
Architecture played a part in it, of course, but the big jump in clocks and efficiency was credited to TSMC 7nm.