Hector Martin

Apple has played the M1 game so damn well it's really hard to tease out the core performance improvements. Still hoping someone does a deep dive into instruction timings, memory bandwidth, latencies, energy usage, etc.

Right now everyone is drooling over insane battery life (how much of that is perf/W and how much of that is power management improvements?), things like glitchless resolution switching (I think many modern GPU pipes can do that... if your drivers are up to par)

Things like unified memory (already a thing on x86 game consoles for 2 generations, e.g. PS4, even though legacy PC drivers on the same hardware have trouble doing it properly).

I've even heard a claim that "the M1 uses less RAM because object acquire/release is faster" going through some nonsense comparison with Android/GC-based systems which, uh, no? macOS has always used reference counting, this just means it's faster, it doesn't use less RAM.

Then Apple also bought their way to being first to 5nm, so that instantly gives them another perf/power boost over the competition (decent leg up on AMD and Intel is left way behind), so that's another confounding factor.

Rosetta 2 is somehow simultaneously amazing new translation technology (it isn't, it's just a really good JIT emulator) but also proof that the M1 is insane because it can beat previous Intel chips Apple used for that range under emulation...