Exactly - Apple hardware is designed for its software, and vice versa. They get battery gains across the stack.

I remember when the M1 Macs first came out, an Apple engineer revealed they'd optimized the hardware so one specific low-level operation macOS does all the time was 5x faster than on Intel [0].

[0]: https://daringfireball.net/2020/11/the_m1_macs

> Apple hardware is designed for its software, and vice versa

Sometimes I find it hard to believe that Apple Silicon implements 2 different togglable memory models, just so that Rosetta can better emulate x86

https://www.sra.uni-hannover.de/Publications/2024/wrenger_24...

It’s not even a particularly obscure low-level operation: atomic add. Every computer in the world performs that exact instruction a huge number of times running normal, non-Apple software.

The key insight is the kind of “vertical integration” providing the kind of feedback loop to spot the opportunity.