I would like to play devils advocate here: What is Volta or Maxwell a "master of"?
-Maxwell can do gaming best (As long as the game doesn't use all of Vega's features)
-Volta does AI best
That's it lol. So if Vega is a master of "few", it's better than any other architecture as far as I can tell. All of the others pretty much only do one thing well.
Not sure what you're smoking, but...
- Versatility. The performance is quite a bit higher in each respective segment (AI/machine learning/gaming/prof. apps) In fact a Pascal
Geforce card already gets similar or better perf in most of the workloads its NOT optimized for versus a Vega alternative. When we get into the Quadro world... its not even a competition
- Perf/watt. The higher performance level is attained within a much tighter TDP budget.
- A much wider boost range. This is why Nvidia cards get put into laptops a lot more. Its 2018 and AMD still has a completely silly GPU boost that is more than a generation behind reality. Vega is the perfect example of it; it boosts itself out of the comfortable temp range and underclocking/volting gets better results. But even then, the boost range remains limited and its one of the reasons the perf/watt stays behind. Its also the reason you don't see Polaris in throttling laptops - performance would nosedive.
Vega only competes in one area and that is price. Price does not make it (GCN) a good architecture. Nor does it finally make AMD turn a good profit on their GPU division - something that has never happened... ever.
You also seem to forget that all of these Nvidia 'architectures' are not some unicorns that all live in separate worlds, they're all branches of the same underlying tech/arch. Nvidia diversifies without losing the edge in any segment, with new systems or tweaks. Tensor cores, NVlink, perhaps a small subsystem for RTX when it gets there... Not with entire new 'architectures'. So yes, the jack of all trades master of none does apply to GCN and does not apply to whichever Nvidia 'arch' you pick. I mean its not even a discussion or a contest anymore. Its a different way of building GPUs and its evident which of the two is more effective.