The M1 Macs

Tuesday, 17 November 2020

Better necessarily implies different.

That’s one of my favorite axioms. It’s both intuitively and undeniably true. But it gets to the heart of a certain universal cognitive dissonance inherent in mankind. We yearn for everything to be better. Yet we are resistant to change.

Apple’s new Macs based on the M1 system on a chip, the first Macs based on Apple Silicon, are that sort of mind-bending better. To acknowledge how good they are — and I am here to tell you they are astonishingly good — you must acknowledge that certain longstanding assumptions about how computers should be designed, about what makes a better computer better, about what good computers need, are wrong.

Some people will remain in denial about what Apple has accomplished here for years. That’s how it goes.

None of these Macs look different from their Intel-based predecessors. I’ve been using an M1 13-inch MacBook Pro for the last week, and it is all but identical on the exterior to its Intel-based counterpart. Same exact size and shape, down to the millimeter. (The M1-based 13-inch MacBook Pro is 0.1 pounds lighter than its Intel-based counterparts, however.) Same keyboard (it’s good), same trackpad, same aluminum color options (Silver and Space Gray).¹ Same exact display. Apple never went for those “Intel inside” stickers, and they’re not changing their stance for the M1 just because it’s their own chip. There’s no marking anywhere on the exterior to tell this apart from an Intel MacBook Pro other than the very small print product number on the underside: “A2338”.

Which is to say that the most exciting new Macs from a technical perspective since 1994 don’t look new at all. You either know that the A2338 model number designates an M1 MacBook Pro or you have to turn it on to know it’s different.

I call these the most exciting new Macs, technically, since 1994 (when the first PowerPC Macs shipped, marking the start of the transition from Motorola’s 680x0 architecture), not 2006 (when the first Intel Macs shipped, marking the start of the transition from PowerPC) because while those Intel Macs were exciting insofar as they were new and completely different for the Mac, they were not different at all, system architecturally, from other Intel-based PCs. Intel-based Macs would never be better or worse than the PC state of the art. Strategically, it was a bet that Macs could be sufficiently differentiated by the MacOS software platform and Apple’s hardware designs around the base Intel system architecture.

It was the correct bet, clearly. Intel-based Mac sales started strong and have stayed strong ever since. The Mac gained converts switching from Windows then, and continues to gain converts from Windows now. In its most recent quarter — the last quarter ever during which the only Macs sold were Intel-based Macs — Apple booked more revenue from Mac sales than it ever had in any previous quarter in company history, and it wasn’t even close. The Mac has never been more popular. The Mac’s previous two chip transitions, from 680x0 to PowerPC in 1994 and from PowerPC to Intel in 2006, were made out of necessity. The old architectures, in both cases, had fallen hopelessly behind the Intel/x86 state of the art.

The logic behind Apple’s transition then to Intel boiled down to another old axiom: If you can’t beat them, join them.

The logic behind Apple’s transition now to Apple Silicon is this: If you can beat them handily, do it.

The M1 Macs are such better machines than their Intel-based predecessors it’s hard to believe. Apple’s battery life braggadocio is warranted. The battery just lasts and lasts and lasts. I’ve been using this MacBook Pro almost exclusively on battery power all week, doing both all my normal work and running benchmarks and performance-stressing tasks, and I can’t come close to depleting it in a full day of work.

It never gets hot. In normal use, it doesn’t even get warm. Maybe, sort of, when running a fully-taxing test like the Cinebench multi-core CPU benchmark, it heats up to just past room temperature above the Touch Bar, but it bears no resemblance thermally to a taxed Intel-based MacBook Pro.

It has been noted that the primary performance difference between this M1 MacBook Pro and its sibling MacBook Air is thermal headroom — that the MacBook Pro has a fan, and the MacBook Air does not, thus allowing the MacBook Pro to run faster longer. The two machines are equivalently fast in a sprint, but the MacBook Pro will win a distance race. But Apple doesn’t call it a “fan”. They call it an “active cooling system”. That sounds like a marketing euphemism, but it’s not fair to call this a “fan”. It is something else altogether, and nothing at all like the cooling systems in any previous Mac laptop.

I’ve never once heard it in an entire week.

Never. Not once. Not a whisper.

I presume it has engaged at times when I’ve taxed the system. Again, the Cinebench multi-core CPU benchmark taxes every available CPU core for 10 minutes. But if the active cooling system has kicked in, I never heard it. I’ve never felt any air moving out of the vents, either. This is nothing at all like the fans on Intel-based MacBooks, which, if you’re not familiar, you can definitely hear, to say the least.

Apple, in its keynote last week, emphasized that the M1 MacBook Air has no fan. (Intel-based MacBook Airs most definitely do. The defunct 12-inch no-adjective MacBook was Apple’s only fanless Intel Mac.) Apple’s point there was to brag that the M1 runs so cool that a high-performance MacBook could be designed without one. Some Mac users, I think, mistakenly took this to mean that the Air had an advantage over the M1 MacBook Pro, in that the fanless Air would always run silently, if sometimes slower. I think this assumption was wrong: the M1 MacBook Pro is, to my ears, always silent as well. Whatever its active cooling system is doing, it isn’t making even a whisper of noise.

No Intel-based laptop with vaguely comparable performance to these machines can possibly match that silence. If you care about noise, the game is already over.

A telling example: I configured this machine using Migration Assistant,² porting my user account from a latest-and-greatest 2019 8-core Core i9 16-inch MacBook Pro using Thunderbolt 3. The 16-inch MacBook Pro’s fan ran the whole time, and it clearly got hot. The M1 MacBook Pro remained silent and cool. Just transferring data over Thunderbolt.

After running Migration Assistant, you know that “first login” experience using a new Mac or after a fresh major MacOS upgrade, when a bunch of long-running background processing tasks are running? The Photos app analyzing your entire library, Spotlight indexing your startup volume, etc. Those are tasks that have always been obtrusive, at least on MacBooks. If they didn’t make your MacBook feel slow to use, they at the very least warmed it and made fan noise. On the M1 Mac, the only way I could tell they were running — despite consuming over 200% CPU for an extended stretch — was looking for them in Activity Monitor.

I have a latest-and-greatest Intel MacBook Air here as well, a review unit from earlier this year. It got hot and the fan noisy just running Software Update to go from the last developer build of MacOS 11 Big Sur to the public release version.

The M1 benchmarks are impressive. Geekbench, Cinebench, Speedometer. Real-world tasks like compiling complex projects in Xcode, encoding video, running Pixelmator Pro’s ML Super Resolution image scaling feature. All of it runs impressively fast and yet the machine remains cool and utterly silent. I’ll leave detailed reports on the numbers to other reviewers, who, I’m sure, will conduct more thorough benchmarkings than I would.

What you need to understand is that the best aspects of these Macs aren’t benchmarkable. It’s about how nice they are. The cooling system never making any noise doesn’t show up in a benchmark. I suppose you could assign it a decibel value in an anechoic chamber, but silent operation, and a palm rest that remains cool to the touch even under heavy load, aren’t quantities. They’re qualities. They’re just nice.

It’s like trying to “benchmark” how nice a trackpad feels. Or the keys of a keyboard. That’s why Apple’s years-long butterfly keyboard fiasco was so utterly incongruous. The defining characteristic of a Macintosh computer is that it’s nice. Beautiful in the way that an elegant mathematical proof is. Apple making keyboards that were very much not nice at all was as odd as if a company whose hallmark was raw speed made something slow. One can’t imagine a single Porsche that struggles to hit the speed limit, let alone a years-long saga of Porsches afflicted with the same deficient engine. But that’s what Apple shipped niceness-wise with its butterfly keyboard.

That keyboard saga is instructive, because niceness, unlike speed, is fundamentally subjective. When your North Star can be objectively measured, you can’t bullshit your way out of a failure. Slow is slow. But when your North Star is subjective, bullshit plays. We can fix it with new materials. We can fix it with a membrane to keep particles out. Less key travel is actually better. These keyboard complainers are just a bunch of malcontents. Bullshit. Forget about the fact that the keys were prone to getting stuck or even falling right the hell off the keyboard, the fact is even when those keyboards were working exactly as designed, they weren’t nice. That’s as much a failure for Apple as Porsche shipping cars that aren’t fast. You just can’t put a number on it.

Conversely, when niceness hits — or, as in the case of these M1 Macs, exceeds — its mark, there’s no quantifiable number to prove it. You just know it when you see, feel, and hear it. (Or in the case of the M1 Macs’ cooling systems, don’t hear it.) Apple knows this. The most telling moment during the M1 keynote was in Craig Federighi’s segment, about 14 minutes in — the bit that launched a thousand memes. Talking about the fact that MacOS 11 Big Sur is Apple’s first version of MacOS ever designed hand-in-hand to run on hardware designed from the ground up by Apple, Federighi’s example was ... how fast M1 Macs wake up.

Not compiling code. Not encoding video. Not executing complex ML models.

Waking from sleep.

It’s not like “my Mac wakes up too slow” is a common complaint. This isn’t fixing a problem — it’s making something that has long already been nice on Macs even nicer. This is the Apple way. The Macintosh way.

Compiling code, encoding video, executing complex ML models — they’re all faster too. But the example Apple went to first was purely about being nicer. No other company in the world would choose that example, because no other company gives a shit.³

Memory, and the Advantages of a Bespoke Hardware Architecture

One thing causing a bit of consternation is that the M1 Macs all max out at 16 GB of RAM. That was true of all Intel-based MacBook Airs, too, but Intel-based 13-inch MacBook Pros go to 32 GB, and 16-inch MacBook Pros and Intel-based Mac Minis go to 64 GB. (iMacs go to 128 GB, iMac Pros to 256 GB, and Mac Pros to, I shit you not, 1.5 TB.)

So what gives? Well, these M1 systems are replacing Intel-based Macs that never had more than 16 GB of RAM. This is one reason why Apple no longer sells Intel-based MacBook Airs — the M1 MacBook Air obviates them all — but is continuing to sell Intel-based 13-inch MacBook Pros and Mac Minis. For now. Apple replaced the 2-port Intel 13-inch MacBook Pros with 2-port M1 MacBook Pros; the 4-port higher-performance 13-inch MacBook Pros (that go to 32 GB of RAM) are still for sale because they’re different models. Their time is coming.

But I think it’s also the case that for most tasks for most people, these M1 Macs need less RAM to perform equivalently to, if not better than, their Intel-based counterparts. This sounds like bullshit but it’s not. It’s the result of a hardware system architecture whose design aligns with Apple’s software architecture.

First, an intriguing benchmark from David Smith, an engineer at Apple:

Fun fact: retaining and releasing an NSObject takes ~30 nanoseconds on current gen Intel, and ~6.5 nanoseconds on an M1

… and ~14 nanoseconds on an M1 emulating an Intel.

Retaining and releasing an NSObject is a low-level operation that is foundational to Apple’s programming frameworks. Just about everything is an object, and when an object is being used, software retains it, and when it’s done being used, the software releases it. This is reference counting in a nutshell: every time an object is retained, its reference count increments. Every time that object is released, its reference count decrements. When it goes to zero, the system frees the object from memory, trusting that it is no longer in use. Retain and release are tiny actions that almost all software, on all Apple platforms, does all the time.

How this works and why it’s so much faster on Apple Silicon than Intel is fascinating but beside the point. The point is that Apple’s philosophical reliance on this model of software development is not typical in the broader world. This is the way Apple thinks software should work, dating back to the origins of NeXTstep in 1989. The Apple Silicon system architecture is designed to make these operations as fast as possible. It’s not so much that Intel’s x86 architecture is a bad fit for Apple’s software frameworks, as that Apple Silicon is designed to be a bespoke fit for it.

Native code running on Apple Silicon is not 5 times faster than on Intel, generally, nor is Intel software running under Rosetta on Apple Silicon twice as fast as on Intel. But retaining and releasing NSObjects is so common on MacOS (and iOS), that making it 5 times faster on Apple Silicon than on Intel has profound implications on everything from performance to battery life. It’s almost silly how much faster this is natively, and quite remarkable that it’s twice as fast even translated through Rosetta.

Broadly speaking, this is a significant reason why M1 Macs are more efficient with less RAM than Intel Macs. It’s the combination of software and hardware designed together. You don’t have to take my word for this — this, in a nutshell, helps explain why iPhones run rings around even flagship Android phones, even though iPhones have significantly less RAM. iOS software uses reference counting for memory management, running on silicon optimized to make reference counting as efficient as possible; Android software uses garbage collection for memory management, a technique that, whatever you think of it philosophically, requires more RAM to achieve equivalent performance.

Now that Macs are moving to Apple Silicon, they can take advantage of the same thing. Cross-platform benchmarks show that Apple Silicon is faster than Intel chips, period. But Apple Silicon’s general performance advantage is increased by running software written using Apple’s platform-specific frameworks. If Bootcamp ever comes to Apple Silicon, Windows likely won’t see quite the same performance gains as MacOS on the same hardware. Likewise, if it were possible to install Android on an iPhone, it would clearly suffer to some extent from having less RAM than designed-for-Android flagships do.

Further, I/O is faster on the M1 than on Intel-based Macs. So when you do run out of free memory and MacOS needs to use swap (virtual memory), it’s faster than ever. It’s not magic, of course. If you really need more than 16 GB of RAM, you need more than 16 GB. But faking it is a lot more fun than it used to be. If you’re using more than 16 GB of memory because you have a lot of small memory allocations that add up to more than 16 GB — like say, a slew of open browser tabs, a dozen productivity apps like Mail and Notes and Photos and so forth, a big dumb sloppy Electron app like Slack, and even something like Xcode, developing a typical application — you really might not notice any hitches with just 16 GB of RAM in these Macs. I don’t. I’ve got all of the above going on right now — plus Fantastical, Tweetbot, Messages, NetNewsWire, MarsEdit, a 12 megapixel photo open for editing in Pixelmator Pro, and of course BBEdit, in which I’m writing this. Activity Monitor reports I’m using just under 7 GB of virtual memory. (This review unit has 16 GB of actual RAM.) Nothing feels any slower or different than when I have just one or two of these apps running. Hell, I just ran Geekbench’s CPU benchmark again, right now, with all this open, and its results were just a hair slower than what I got when it was the only app running.

I’m not saying 16 GB ought to be enough for everyone. If you’re doing work involving discrete chunks of data that themselves are multiple gigabytes in size, you need many gigabytes of actual memory. But a lot of small things that add up to more than the sum of physical memory on an M1 system does not notably slow it down. As a well-known hoarder of open browser tabs, I was skeptical about “only” having 16 GB of RAM in a new MacBook. But having spent a week deliberately trying to tax this system’s memory limit, I’m a believer. I wouldn’t hesitate to buy one of these M1 MacBooks today — at least the 16 GB configuration I’m testing — with the expectation of using it merrily for years to come.

It’s possible to make good computers with general purpose hardware and general purpose software created by completely separate teams. One group makes hardware that tries to be good for any sort of software. The other makes software good for any sort of hardware. That’s how many enthusiasts think good computers have to be made, simply because that’s how it’s usually worked. But such systems will never be as good as hardware and software designed hand-in-hand together, blurring the lines between which side is optimized in which ways to suit the needs of the other.

Battery

As I type this paragraph I’ve been working for just over three hours, nonstop, with the MacBook Pro unplugged the whole time, and the display as bright as I want it to be. The battery is at 80 percent. To say that it offers merely “all day battery life” would require me to work very long days.

Rosetta

Rosetta is a marvel — the exemplification of “it just works”. x86-compiled apps run just about as fast as they do on the most expensive Intel-based MacBook Pros, and faster, by a long shot, than on the last and best Intel-based MacBook Air. Let that sink in: apps compiled for Intel run faster in translation on the M1 than they do on actual Intel CPUs in MacBook Airs and most MacBook Pros. Here, fine, I’ll show a few benchmarks from Geekbench (single-core / multi-core):

• 16-inch MacBook Pro (2019, 8-core Core i9): 1,160 / 7,160
• MacBook Air (2020, 4-core Core i5): 1,050 / 2,410
• M1 MacBook Pro (x86 benchmark running in Rosetta): 1,260 / 5,600
• M1 MacBook Pro (native M1 benchmark): 1,730 / 7,530

You just don’t have to worry about Rosetta, period. The first time you launch an Intel app, MacOS will ask you if you want to install Rosetta. Agree, and installation flies by in a jiff — mere seconds — and after that, you’ll just never notice whether something is compiled for Intel or native Apple Silicon. You don’t have to worry about extensions (like for Safari or Finder) being matched — Intel extensions run just fine in Apple Silicon apps. Dropbox, for example, is still an Intel app but it runs just fine, including Finder integration.

Rosetta performs great and compatibility is seemingly seamless. This is a technical tour de force, a home run for Apple that will mostly go unheralded by typical users because the entire point is that they shouldn’t notice or care.

The Camera

The one spec on these new MacBooks that elicited a universal groan was the “720p FaceTime Camera”. 720p is the same resolution as Apple’s Intel-based MacBooks, whose cameras are universally regarded as craptacular. You can’t judge any camera by resolution alone, though, and Apple is billing the camera systems in the M1 MacBook Air and Pro as being improved by dint of the M1’s image signal processing. Quoting from the MacBook Pro product page:

Thanks to M1, the FaceTime HD camera can now take full advantage of our latest image signal processor — improving image quality in video conferences and pulling out more details in both shadows and highlights.

Is the camera quality very noticeably improved? Yes.

Is the camera quality now good? No.

MacBook FaceTime camera image quality was so bad that there was plenty of room for the M1 MacBook camera to get a lot better and yet still be described as “not bad” at best. Here’s an A/B comparison from a FaceTime call with my friend Adam, conducted at night, with no attempt whatsoever to adjust the lighting in my office in a flattering way. One shot is from a 2019 16-inch MacBook Pro, the other from the M1 MacBook Pro. The difference is striking, even if the M1 camera remains on the whole unsatisfying. (I look oddly pleased, if not downright eager, in Adam’s capture of my image from the M1 camera, and I will get him back for this someday.)

iOS Apps

MacOS 11 “Big Sur” introduces one major new feature exclusive to Apple Silicon Macs: the ability to run iPhone and iPad apps from the App Store.

This sounds fine on paper, but in practice I don’t understand who thought this was a good idea to ship. My experience has ranged from terrible to OK, at best.

The best experience I had was using Marco Arment’s excellent Overcast podcast player. Everything worked, including syncing with my account, and Overcast’s window, when running on the Mac, is resizable like a Mac window should be. Overcast opens by default in a pseudo iPad-size layout, but I prefer using it in a tall skinny window, more like an iPhone single-column layout.

The gaming experience was meh. Games that I consider a ton of fun to play on my iPhone worked just fine but were no fun at all to play on my Mac. In the same way that games designed to work with a hardware game controller never feel right without one, games designed to be played on handheld touchscreen devices are no fun to play on a laptop.

The most inexplicable experience was HBO Max. HBO Max running as an app on MacOS is worse in every single way but one than using HBO Max’s website in Safari. The window is, for a video streaming app, tiny, and cannot be resized. Nor does full-screen mode work. You can watch video in this tiny little window or not at all. Scrolling lists with the trackpad feels like jabbing at a dead fish to slide it across a countertop — scrolling has no momentum or bounce or just plain life to it because I guess HBO coded the whole thing up in some shitty cross-platform framework. Just try getting down to “Xfinity” in the list of cable providers for signing in. (Scrolling lists via the trackpad in Overcast works just fine, full of life and momentum, because Overcast is a proper UIKit app.) The one and only thing HBO Max has going for it as an iOS app running on MacOS, that you don’t get from the website, is the ability to download videos for offline viewing. That’s it. But why even bother when you can only watch them in a tiny postcard-sized window?

It’s possible HBO will fix some of this. Just making the window resizable and enabling full-screen video playback would make the app at least useful. But even at best, like Overcast, iOS apps running in a window on a Mac feel foreign. They feel like what they are: apps from another platform. I can see how some people might think this is a good idea, but I don’t see how anyone thinks it’s a very Apple-y idea. Sure, it works, which is why most companies would just ship it. More apps are better, right?

But they’re such a crummy experience, these iOS apps. At the very best, sub-optimal. This feature exemplifies a spirit of “better than nothing, ship it”. The Apple way, typically, is “insanely great”. It’s like someone said, “Oh, you thought lazy Catalyst ports were a bad experience on MacOS? Hold my beer…”

Each of these apps has an option in the application menu: Touch Alternatives. The first time you choose that menu item, you’re presented with the following alert:

The ugliness of this dialog — they didn’t even capitalize “Option” for chrissake — speaks to the experience of using this feature as it stands today.

Conclusion

Folks used to say that no one gets fired for buying IBM. Similarly, no one ever won by betting against Intel and the x86 architecture. Intel always won.

Until now, that is, when it’s clear they’ve lost. Apple’s two previous Mac architecture transitions were made from positions of desperation — they were jumping the Mac from sinking ships. This transition from Intel to Apple Silicon has clearly been made from a position of overwhelming strength. Yes, we still haven’t seen the Apple Silicon chips for high-performance MacBook Pros, let alone for desktop workstations. Many technical questions remain for the remainder of this transition. But the M1 is a rousing start. Apple surely could have initiated this transition a few years ago, but they waited until the effect would be colossal, the advantages undeniable. I chatted with one Apple employee who’d been using this hardware for months, and after it was unveiled, his daughter texted him to ask if the new MacBooks were faster than hers. “Much” was his one-word response. Then he texted again: “Much much.”

You can add as many muches as you want, and it won’t convey the seismic shift the M1 represents. No one talks about not getting fired for buying IBM anymore. Soon, no one will think you always lose betting against Intel and x86.

Think different, indeed.