Starting small and going large is not a new concept.
Nvidia did exactly that with Maxwell. The first Maxwell parts were the 750 and 750ti despite the rest of the 700 line being Kepler. The 900 series followed about 8 months later with Maxwell proper.
It worked well for Nvidia I can see it working well enough for AMD too, but I’m sure there is a reason Nvidia isn’t still going that method. Or maybe there isn’t Iunno.
Good point and I did honestly forget about NVs maxwell parts.
I think for NV the last couple generations (not 1000) but Volta and Turing. They have tried to have their cake and eat it too. Saving a ton of money designing one chip to rule them all. Now with no real high end competition from AMD perhaps it wasn't a bad bet. They saved the cost of designing and tapping out an actual consumer part for a long time by simply selling their AI/Server designed silicon cast offs into the consumer market with RTX.
IMO They choose great profits... by salvaging all those chips they have been able to compete very aggressively on the AI end of things, and they have been pretty safe on the Consumer end jacking up the pricing as AMDs Vegas where never able to really match them in performance and has issues with heat and power.
I think NVs going big and big only was probably a pretty smart move really. I think it has only worked because AMD sort of dropped the ball. Vega wasn't as good as it should have been and the choice of HBM memory made it almost impossible to bring Vega costs down. So AMD has been selling very old polaris stuff for a lot longer then anyone would have imagined.
I do hope NV is going little... then big with Ampere. I think from everything Intel has said publicly so far it sounds like XE will also be a little... big roll out. With consumer cards differing quite a bit from the high end XE server parts.
The CPU fight got interesting in 2019... it seems like 2020 might be a good year for GPUs.