The problem is that those same "holy grails" numbers shouldn't be repeated over and over: you know that an architecture/process isn't just x% more efficient than another. It's x% more efficient at a certain speed, temperature, ecc.
Comparisons are done according to scientificals protocols, wich mean that the tests conditions wont favour a design over the other, here we are compairing close designs so the question doesnt even exist.
The
So at least stop using a single value that surely isn't always the correct one, also... why Skylake couldn't be much more efficient on the same process?
I gave an estimation of 10-15% better perf/watt for BDW in respect of HW with only very variable laptops reviews and well before Hardware.fr posted their DT review, susprisingly the DT numbers perfectly match the mobile SKU numbers apparently, and this say that if Intel process is mediocre relatively to the previous one its linearity is still very good, much better than what AMD has at hand.
Isn't Carrizo built on the same node yet better than Steamroller?
No and yes at the same time, this is the same process but with smaller transistors that have badly degraded conduction but also require less drive energy, this allow lower leakage, lower dynamic power in exchange of less frequency.
Both architectures are literally built for mobile and better efficiency/performance, so I hardly see the problem here...
Indeed Intel made the same choice, that is transistors with degraded conduction but with same advantages as above, what is surprising is that AMD made a much better work in this respect given the node disadvantage, and this allow them to get exceptional perf/watt below a given TDP that is at 25-30W, Intel is better above but not below, and only for the CPU part.