Core U models are known to blow it's TDP worth of power consumption in short duration workloads until they reach certain temperatures (targetting conviniently benchmarks) by about 2x.
Core M models do the same, but by x3, but the duration of the boost clock is inconsistent with the OEM's cooling capabilities. No wonder they are nearer to the U models in performance than what their TDP would suggest.
AMD implemented this too with Carrizo, but the TDP worth of power consumption is only exceded by 1/2 to 1/3 of its nominal TDP, and it is time based in second and not temperature based. Moreover, you can lock the power target of these chips to avoid this behavior, resulting in more consistent results when benchmarking.
This obviously doesnt mean you wont be seeing posts like the one I quoted when uninformed people take Intel's TDP values as strict power consumption measurements and make such bizarre claims like Intel is "blowing" AMD 15W chips with their U and M lines. Those chips indeed are consuming 15-20W at most in benchmarks but Intel Core U ones are consuming roughly 30W and Core M ones 15W. So the poster that linked benchmarks of the same AMD chips but with power targets set at 35W make more of a fair comparisons againts the U line, while the 15W target results are useful to compare against the M line.
So without even knowing, the poster I quoted admitted that at same actual power consumption, AMD Carrizo and Intel's Core line are trading blows in overall performance (with Intel winning in ST and AMD making it up in MT, while on iGP department AMD comes on top when dual channel is used). The problem is, one company has their product in 14nm FF (second gen), while the other one is using Glofo 28SHP.
My take on this AMD should go full Intel and make their chips overblow their TDP values as aggresively as Core products do, after all people wont ever care as shown by the level of ignorance regarding Intel's own power consumption numbers.