Eh, a cherry-picked card sent to a compensated reviewer might be a reasonable guess. Wouldn't be the first time. The box was open when he got it, and he stated he was asked to review it, etc.
But remember, power consumption in clocked circuits is related to the voltage squared, so even a small reduction in voltage can make a big difference in power.
I know this perfectly. The voltages are there, in the video displayed while benchmarking. Standard voltage under load for this GPU: 1.05v. Reference Voltage on reference RX 480 - 1.25v. Both in boost core clock cases.
There is a problem. Reference designs do not clock that high up on that low voltages. Remember, this GPU has 1288 MHz@1.05v vs 1266 MHz@1.175v reference. Proofs? Here you go:
The only thing that could make it so is that the GPU is from new revision of silicon.
Look at the voltage curve and Frequency curve. And compare that to completely abstract 1.175v and 1475 MHz. Possible with previous versions of Silicon?
And to add to that all of this: Embedded versions of Polaris GPUs have 95W TBP(Typical Board Power) at completely stock, reference 1266 MHz but unknown voltage for Polaris 10 GPU, and under 50W TBP, for 1.4 GHz(!) Polaris 11 GPU. And RX460 which uses Polaris 11 die rarely goes down below 80W under load at 1.2 GHz core clocks.
So something must have changed, lately.