Apple's IPC advantage isn't hypothetical; it's been measured (mostly by AnandTech) for years.

That being said, I'm a little skeptical of a 4 big + 4 little A14X beating Intel's 8 big cores.

The time it takes to execute an x86 instruction is not constant and the amount of concurrently executing instructions (pipelining) is not the same between these two CPUs. Also, power management works every different in arm so it has a lot of boosting head room.

Not too familiar with cpus but if it's known that arm outperforms x86 while also being more power efficient, why aren't everyone using it? Why did it take until now for a major computer manufacturer to switch?

ARM doesn't outperform x86. Apple Silicon—which only Apple has—outperforms x86, and only this year.

ARM did outperform x86 at the start. An Archimedes could do floating point faster than a Compaq Deskpro 386 with a 387 FPU.

> f it's known that arm outperforms x86 while also being more power efficient, why aren't everyone using it?

PowerPC and Alpha did that in the 90's and the reason is software.

Intel's x86 has a huge software ecosystem. In 1999 (thanks to a procurement mistake) I was reading my e-mail and browsing documentation on a wicked fast 64-bit computer. It would be a good couple more years before I could edit a Word document on a 64-bit computer because Microsoft didn't have Office for them. I only got to read e-mail on that ridiculously fast Unix workstation because nobody wanted to use it and it was just sitting there running our timesheet software and magnificent screensavers.

Multi-processor x86 desktop machines only became popular when Windows XP became popular. Multi-processor Macs only became popular when Apple moved to OSX. IBM only added multi-processor support for their mainframes when they hit a dead end and had to move to machines with slower CMOS processors with more processors to make up for the speed they lost.

Software dynamics holds hardware evolution back.

Part of the reason is that ensuring things work for a different CPU architecture is simply a non-trivial task. There are libraries with hand-written x86 assembly, and others in C/C++ that depend on x86 intrinsics to perform acceptably, and then there are those that work accidentally on x86 because of how compilers implement some undefined behaviour for x86.

Cloudflare did a very interesting post that made the rounds here: https://blog.cloudflare.com/neon-is-the-new-black/

Discussion: https://news.ycombinator.com/item?id=16831609

It's not always just instructions-per-clock, sometimes it's also data-per-clock. Specifically, the Apple Neural Engine in the Bionic chips can perform some vector computations faster than anything in either the CPU or GPU of any Intel Mac.

Note that it says it outperformed an "Core i9 MacBook Pro" not a "Core i9" so depends on throttling due to heat. It's quite easy to believe the A14X is more efficient and can win in those conditions.

Geekbench is a scam