The fact is that x86 has 4 decades of backwards compatibility which along with the PC architecture means a reasonably open and documented platform. Every ARM system is different and they are all lacking the heritage of public documentation that the PC has. In other words, it's not about performance but freedom. I doubt Apple will ever release as much detailed information on its own chips as Intel and AMD have.
x86 is plenty fast anyway, and still wins on code density in single-threaded scalar (i.e. NOT using the vector units) branchy code of the sort commonly found in average applications.
Essentially all the supposedly merchant SoCs are proprietary. Instead of shipping data sheets, winbond or mediatek want to ship you a reference design with reference linux or android build that only works because of a huge binary blob.
There’s a small hardware consulting company in Palo Alto whose main differentiator is that they have a bunch of folks in China who manage to generate docs they can use. I’m guessing this involves a lot of under the table leaks and late night alcohol-fueled discussions.
Thanks, so it's really about drivers etc rather than the architecture, which makes a lot of sense. A bit surprising that there hasn't been more convergence in the Arm world.
x86 is plenty fast anyway, and still wins on code density in single-threaded scalar (i.e. NOT using the vector units) branchy code of the sort commonly found in average applications.