Hector Martin

But what about general purpose computing? I mean sure, it's going to beat Intel on performance/watt (anyone can do that lol) but what about peak performance? The only benchmark that stands out is "Build code in Xcode up to 2.8x faster." but how much of that is CPU vs SSD?

I'm still standing behind my prediction that Rosetta 2 is going to disappoint for number crunching, which is what people in any plug-in ecosystems are going to be stuck with for many years to come.

Now the other question is whether Apple can deliver a true high performance SoC to compete with the x86 solution on the high end. M1 obviously isn't it. 16GB max and no eGPU/dGPU support means it can only compete with ultraportable/embedded class x86.

Now if they can come up with something that competes in the same class as a PS5 SoC (to give a baseline x86 design by a competent company, because Intel have been failing for years), *then* things will get really interesting. But today is not that day.

What we're seeing right now is Apple stretching the mobile/tablet market segment hardware, which they've gotten *very* good at, to encroach on Intel's ultraportable/embedded segment, which they've sucked at. Logical. Now can they go higher? We'll see.

Oh yeah, and One More Thing. Rosetta 2 *necessarily* uses more RAM than native execution. You need to keep the ARM JIT/AoT code/caches *and* the x86 code around, and probably some overhead on redundant JITting too etc. Now consider the 16GB limitation on top of that.

Very smart people have been doing high-perf emulation (yes it's emulation, *every* modern emulator uses dynamic translation and has for over a decade) for a long time. There is no such thing as "transpiling an x86 app into ARM". It's just not a solvable problem.

Oh there is look at @_revng