The New CPU ARMs Race | Hacker News Reader

The New CPU ARMs Race

progbytes.substack.com

11 pointssudhirj6 comments

Nit: this article keeps talking about "The RISC" and "The CISC" as if there are only 2 instruction sets in the world, rather than appropriately naming them the x86 instruction set(s) and ARMx instruction sets that happen to be CISC and RISC.

It also makes a giant (and incorrect leap) to say that software is why this is only happening now. Aside from the fact that there's no clear reason why 2020 is different in terms of software (what changed?), it pushes aside performance differences that have plagued moving ARM beyond mobile phones for years. Multicore performance, vector/float performance, clock speeds... all these things have been thorns in ARM's side for a long time now while also being a feather in Intel's cap. So either these processors will make a different trade off (SMP workloads are poor, but battery is great), or AWS/Apple have a breakthrough that has changed the game. Apple's internals are secret, so we shall see once the benchmarks (and products) finally come out!

Author here, thanks for the feedback. I want to keep this simple enough for me and my non-hardware engineer colleagues, but will see if I can add a note about the instruction sets.

Yeah, I remember when only the cheap slow devices were ARM, and that suddenly changed. I'm assuming it's because some very talented engineers at Annapurna/AWS and Apple starting working on the problem with heavy iteration – if you look at the Apple A series performance it shows lots of compounding benefits. AWS is much more opaque, but they're likely iterating like crazy internally, and their Graviton2 is a better business decision for customers who can run ARM, and they're layering ML and encryption onto it.

Will fix the note about software being the reason the transition now – I meant to say Apple has had the A series chips for a while, why are they deciding to transition now – probably because it took a while to get the OS and Rosetta ready.

> probably because it took a while to get the OS and Rosetta ready.

I think you're making a big assumption here, and ignoring that shipped ARM chips haven't been tailored for this kind of load until recently -- and the cat's not out of the bag yet for Apple.

The article says "Intel and AMD license the CISC, so they’re the only ones making x86 chips". The "so" in this sentence doesn't seem to make sense since if they're licensing it then presumably others are making them. Also, AMD and Intel aren't the only ones making x86 chips since both VIA and Zhaoxin are also making them (and possibly others that I'm not aware of).

Thanks, will edit.

> Intel, on the other hand, runs on the CISC specification / instruction set

Hmm.

Reply on news.ycombinator.com