[Thread] A look at the “black magic” that lets Apple's M1 chip achieve extraordinary performance and efficiency gains
1/ In case you were wondering: Apple's replacement for Intel processors turns out to work really, really well. Some otherwise skeptical techies are calling it “black magic”. It runs Intel code extraordinarily well. Tweets: @gruber , @jsnell , @ohmdee , @kartones , @newsyc100 , @aeyakovenko , @nfonrose , @xeraa , @erratarob , @mostalive , @drewthaler , @stuartd , @erratarob , @anaisbetts , @bluxte , @codefolio , @pangaea__ , @adriftfuriously , @helgeklein , @davemark , and @charlesarthur Tweets: John Gruber / @gruber : @charlesarthur In hindsight it's kind of amazing that it has taken until now for chip designers to optimize CPUs in ways specific to making JavaScript faster. JavaScript is not going anywhere, and it is a huge bottleneck for a lot of software a lot of people run all day. Jason Snell / @jsnell : @gruber @charlesarthur He's an expert troll, but I do enjoy this tweet - optimizations make M1 Macs “seem” faster, by running JavaScript faster. That's not a use of the word “seem” I was aware of. https://twitter.com/... @ohmdee : Once again, Apple has executed excellently & admirably with their latest silicon efforts. I'm thoroughly impressed with the combination of clever solutions in M1, culminating in a momentous lead. The Mac has re-entered my purchase consideration set for the first time in years. https://twitter.com/... Diego / @kartones : I don't like and no longer need use a Mac for work, but each week I'm more and more interested in the M1 hardware changes and clever approaches to speed up things, between Apple and AMD the mostly stagnant Intel should pick up the pace: https://twitter.com/... @newsyc100 : Apple CPU tricks: memory reordering, JavaScript support, ref counting https://twitter.com/... (http://news.ycombinator.com/ ...) Anatoly Yakovenko / @aeyakovenko : Snapdragon 875 has a 4mb L3, M1 has a 12MB!!! L2. It's not “magic”, Apple just wins by using a ton more die. Apples cost savings from vertical integration are spent on silicon, which results in a product that is really hard to beat. https://twitter.com/... Nicolas Fonrose / @nfonrose : Very interesting thread on various choices Apple made for the M1 processor It also shows how important it is for Apple to control the full stack, from the language (Swift), down to the silicon (contains instructions to optimize the reference counting memory mechanism Swift uses) https://twitter.com/... Philipp Krenn / @xeraa : some good background info on apple's ARM processors and why they are so fast — without black magic https://twitter.com/... @erratarob : @charlesarthur Of course I didn't mean any negative connotations by “cheat”. I mean only that there's convention wisdom about what you shouldn't do, and Apple did it anyway. :) Willem van den Ende / @mostalive : E.g instructions optimized for JavaScript. In the 80s there were processors optimized for Lisp and Smalltalk, but these were behind in other areas because this of low volume. M1 also gets other things right . https://twitter.com/... Drew Thaler / @drewthaler : Great thread on Apple's M1. As the thread says, none of this is black magic, just “all the various things have been executed really well, leading to a combined result that is a great leap forward.” https://twitter.com/... @stuartd : @ErrataRob @charlesarthur As a software dev, I'm wasn't sure how to process “cheating”. It either works it it doesn't. And in this case, it seems to work fine. But us software guys are, well, soft. I'm interested though, if you have a moment, what's the ‘conventional wisdom’ that's being ignored here? @erratarob : @stuartd @charlesarthur The standard way of designing a CPU is for general purpose use. Apple didn't do that — they designed it for their own code and nobody else's. In retrospect, this is obvious of course. Ani Betts / @anaisbetts : A lot of hype around Apple ARM64 chips but this thread digs out the *actually* cool stuff about Apple's new processors https://twitter.com/... Sylvain Wallez / @bluxte : Apple's M1 processor has a number of tricks. The most clever one for the x86 transition is certainly the ability to emulate its memory ordering modes. https://twitter.com/... Noah Gibbs / @codefolio : Okay. I had been skeptical about M1 Macs. These explanations remove my skepticism entirely. These are *good* tricks, but they are also *achievable* tricks. Also, it's not that M1s are massively fast and efficient. They are massively fast *or* efficient. https://twitter.com/... Pangaea / @pangaea__ : this is actually impressive as hell, cool stuff from apple https://twitter.com/... FuriouslyAdrift / @adriftfuriously : @ErrataRob This is the first Mac since the PowerPC era that's definitely worth the money... I might get one and I don't even like OS X. Well done, Apple... Helge Klein / @helgeklein : This is by far the best info I've seen on the new Apple M1 CPU. Many points should be valid for the A-line CPUs, too. Explicit optimization for JavaScript explains the iPhone's fantastic performance, as does iOS' reference counting vs. Android's garbage collection. https://twitter.com/... Dark Mode Dave / @davemark : Great 🧵 on why the M1 can do the crazy things it do. https://twitter.com/... Charles Arthur / @charlesarthur : This thread seems to 1) imply Apple “cheated” by getting x86 code on ARM to run better than Microsoft managed. Why not say “made a better choice”? 2) imply that optimising Javascript (there's a lot of it on the web) is underhand. No, it's smart. 3) suggest DTKs are sneaky. Huh? https://twitter.com/...