It begs the question, why didn't AMD just go for more cores over higher clockspeed? In the review by Anandtech, they stated that they spoke to an AMD engineer who said they are not limited by 4 compute engines but instead chose to stick with that design and just increase clocks and focus on features such as hbcc . They decided to use 4b transistors and a decent amount of die space just to increase clockspeeds, instead of adding more compute engines. A 6144 core Vega clocked similarly to Polaris 20, i.e. 1300 MHz base and 1400-1500MHz boost would most likely be enough to solidly be ahead of the 1080 and get into 1080Ti territory.
Perhaps they actually see adding more cores to the die as being a dead end for them in future development and didn't want to waste R&D funds on developing it? According to GloFo they expect a greater than 50% reduction in die size, going from 14nm LPP to 7nm LP with a sizeable increase in performance via higher clocks also expected.
http://www.anandtech.com/show/11558...nm-plans-three-generations-700-mm-hvm-in-2018
If this is the case then AMD's design for Vega may in fact be a stepping stone to Navi, which will utilise multiple dies like Threadripper and Epyc. If they can produce the same Vega 10 die on 7nm LP, it'll be comparable in size to Polaris 10 and with the higher clock speed design of the architecture it should be able to boost in excess of 2 GHz with the smaller node size. If they stick four of those dies on an interposer or as a MCM, that'll give a 16000+ core monster which they can clock down slightly to save power whilst having incredible performance as an overall card. Then with the mid range they'll use a single die but clock it high as possible to get the most performance out of it at the cost of power efficiency.
So tldr the reason why Vega performance is disappointing, could be because it's just a stopgap essentially.
megasadface
