No Intel messed up at 14nm and it's obvious when looking at performance per watt metrics. Just look at Haswell vs. Skylake. Although none of the reviews really pointed this out it's embarrassing the Skylake K series chips have a higher TDP than their predecessors which taken into consideration offers virtually no improvement in overall performance. They clealry messed up and I suspect that's why they launched the leakiest K series parts first instead of where they traditionally launch in notebooks or iMacs etc.
I would like to understand your logic behind the whole "scaling down is harder than scaling up" statements. This a nonsense. Chip manufacturers have a transistor budget and target a specific TDP and build the fastest and most efficient chips they can around these limitations. Almost all mobile chips are "Bursty" as they quickly hit thermal limit thresholds due to form factor and battery constraints as well as fabrication limitations, but the advances shown by the OP show they are tackling problem this while still dramatically improving performance.
The big vs small design statement is cop out for Intel apologetics. They messed up and should be called out for it. A 50~ percent increase in performance over 5 years is embarrassing. Maybe this is just a reflection of no real competition but they can't get away with this crap in the mobile market.
That being said Intel does have multiple chip designs, some of which directly address mobile device markets and with new phones like the Asus Zenphone 2 they are clearly making progress. They still have a long way to go before catching up to QC, Apple or Samsung (SoC development is a lot more than efficient chip design) but with enough Gorilla marketing they could make a small dent with a few more design wins.
First I never said the "scaling down is harder than scaling up" that is you putting words into my mouth. You are not understanding what I am saying if you thought my comments can be summed up by that. Scaling up in performance is hard when your best chips are already running at 4 ghz, but if they were only running at 1 ghz it is far easier to scale up. Scaling down in power consumption is hard if you have your high end chips already running at 0.7 volts, and you want to go even lower, but it is easier when your high end chips are at 1.2, 1.3, or 1.4 volts and you want to scale down even lower in voltage.
I can go very in depth into this if you want, but I rather not do so if you already are familiar with enough of the physics and math. Me explaining in full is kinda rambling and I am trying to be short for it is easier for people to follow what I am saying the less words I use. (note this is me trying to be short )
Power Consumption is governed by this formula
P=
C*V^2*F
Power =
Capacitance times
Voltage times
Voltage times
Ffrequency
When you have a 4ghz chip it is harder to increase that performance by 25% vs a 2ghz chip and then increasing the performance 25%, this is due to that formula not being a linear formula. Each time you increase the frequency you need to increase the voltage due to the effects voltage have on Capacitance (clearing the circuit so it is ready to turn on or off depending on what it needs to do next) for the higher the frequency the faster you need to make the capacitance "blank out", and by doing so you hit a power wall for besides increasing your power consumption linearly just by increasing frequency you also have to add far more voltage and voltage adds quadratically and not linearly to the power consumption.
For example desktop chips run their chip at their high stock and non turbo clocks at 1.2 to 1.3 voltage even on broadwell and skylake. Core M by contrast is targeting 0.7 volts and actually runs less than 0.7 at their non turbo base speed of 1100 or 1200 mhz depending on model. Lets pretend there is a 14nm desktop broadwell core i3 on the market, currently there is not and probably will never be with a skip directly to skylake, but lets pretend the Core i3 4370 @3.8 ghz Hawell chip was really a 14nm braodwell.
So if you take the identical desktop i3 and downclock it to 1.2 ghz of the Core M and keep it at the same voltage at 1.2 volts that it would take to run a 3.8 ghz chip vs the 0.7 volts of the 1.2 ghz core m and you turn off turbo the desktop chip will use
1.2*1.2/0.7/0.7=2.93 times more power, even though they run the same clock speed just because you are wasting heat due to voltage. But it is actually a higher number once you restore that chip to 3.8 ghz for now you also have to 3.6 times as much power just due to the difference of 3.8/1.2=3.66. Also the number goes even higher for when you increase the voltage you also increase the temperature due to resistance and this in turn modifies Capacitance physical properties of the material making the chip even less efficent. So 2.93*3.66*Temperature Loss in efficiency means you are using over 10 times as much power plus whatever the loss due to temperature to only gain 3.6x more speed.
Here is a great forum post by a user of this forums called I Don't Care, he is an elite member of this forums but he actually works as a cpu designer as his profesion, but he did a great hobby post showing the power consumption of an i7 2600k and an i7 3770k at various temperatures and voltages to illustrate this fundamental physical equation I showed earlier in real life conditions
It has wonderful graphs
i7-3770K vs. i7-2600K: Temperature, Voltage, GHz and Power-Consumption Analysis
http://forums.anandtech.com/showthread.php?t=2281195
-----
Thus it is harder for intel to make a desktop process faster than it is to make a mobile process more power efficent. It is easier for arm for they are starting at a smaller chip and scaling up.
Though even though it is harder for intel to increase total performance on a desktop chip, it is very easy for them to increase performance by watt by scaling down and making the process more efficient by doing tricks so it uses less voltage, increase capacitance with things such as finfets, reduce leaking, etc. This is just like it is easy for ARM to scale up.
But at the same time it is harder for ARM to get more efficient not on performance but on minimum power consumption for they have been doing that for so long.
Booth companies have challenges but they are purposefully emproaching on each other turfs for you can't fight physics and it is harder for them to be even better at what made them already famous, and thus the low hang fruits is to do the things they have not specialized in.
-----
I am sorry but this is not opinion but flat out physics, electrical properties, chemistry, and math, there is a limit how much you can out engineer the fundamental rules of the universe. Thus looking at performance per watt is a more equal comparison at judging improvements of the cpus over generations, even if you can flat out care less about performance per watt for your desktop like mine does not care if you save 10 or 20 more watts, for it is a 200 to 300 watt machine inside a case and power supply that could handle a 1000 watts if you really wanted to push it. I purposefully did not pick a 1000w psu but instead a 620w seasonic that is fanless unless I play a video game, yet my case can handle 3x or 4x gpu for I picked it due to quietness and easy of assembly and tinkering even though I just use a single gtx 680. It sucks but it is harder to make a desktop cpu faster in single thread, a desktop gpu is different, adding more cores is different, adding a faster ssd or faster ram is different but single thread is very hard to improve today on desktop form factors. It sucks but it is true.
Now getting cpu performance that we used to have in a desktop in 2008 to 2011 in a cell phone / 7 inch tablet form factor is far easier but even that is not easy. I am glad we are now starting to see this from both intel and arm, and if AMD had better fabs and more engineers to spend on fine tuning their cpus we would see this from AMD as well.