Mobile GPUs are not as plug & play as desktop GPUs, partly because you often have to let multiple GPUs work together (integrated + dedicated) and because your cooling and power budget must be matched. Therefore, it is even more difficult to distinguish real variants of the GPU from fake ones.
That problem is also why you still see laptops on the market whose manufacturers release drivers themselves to ensure that everything works properly. Your GPU maker’s official drivers sometimes don’t work or act crazy, and then you have to wait for the manufacturer to release a new driver (good luck with that).
This sounds like a bug in the “wrong firmware on the wrong SKU” category. Can happen, of course. Of course it could also be that the artificial capping was done on purpose to keep the temperature low during testing and get nice results, but if that was the case you can still count on those nice results if you wanted to and risk it they only anger their customers and shareholders when something goes wrong. I’m guessing it was just a mistake.
Manufacturers of mobile GPUs often have integration problems, the name and numbers say nothing about the performance. I recently saw a review of a mobile 3070 that beat a mobile 3080 because the cooling wasn’t overloaded and because more power was going to the GPU instead of the memory (because there’s just a few GB less RAM in those things! ).
As a consumer, different benchmarks from different reliable parties are the only way to distinguish a good GPU from a bad GPU. Nvidia is increasingly letting manufacturers publish the TDP with the GPU, and you can assume that TDP per class actually tells you how good the GPU really is. I don’t know what the situation is at AMD.