And the Rest Of It

Mark II

The Narnian Wardrobe of Juicers

January 16, 2021 · A 5 minute read.

This post is arguably the post that brought this blog into existence. Well, that and Medium becoming an increasingly terrible platform—that was another big factor. But leading up to the announcement of the M1 Macs, I had a lot of thoughts and theories about what kind of performance and battery power these chips would have. I started building a Strapi + Nuxt blog, which took surprisingly little time. I just had surprisingly littler time to spend actually working on it. The blog wasn’t going to be ready in time for the Mac event, so I just tweeted about itextensively.

My semi-educated guesses were surprisingly accurate, honestly, and I was a little surprised. But then, I spent an inordinate amount of time researching this, or at least that’s how it appeared at the time. But the new Macs are out now, and they’ve shown us that Apple has not simply matched Intel performance. They didn’t even just beat Apple performance. They changed the rules of the game and then set a new high score without even really trying. There are a lot of reasons for this, and I’d like to write about all of them at some point, but this post is going to focus on performance per watt, because I don’t think it’s well understood, and I don’t think it’s talked about as much as perhaps it ought to be.

Less Fruit, More Juice

It’s important, at the outset, to understand that performance, power usage, and heat production are all highly related. A computer takes in electricity, which produces heat, and using this electricity it pops out numbers that will turn into cat videos, blog posts, and deep fakes of Nicholas Cage in movies he never should have been in.

The question of performance has always been about how much power we can give a chip and how efficiently it uses that power. This is why bigger, heavier, hotter computers are always more power efficient. This is why if you buy a 17-inch Lenovo Workstation laptop that comes with a power brick you could use to maim someone, it will inevitably outperform something like a MacBook Air or a Netbook (remember those). Or at least, that used to be the case.

Let’s use a metaphor from a field I know nothing about, so I can butcher it and get emails. If you’re juicing oranges—or let’s say apples, I imagine there are two ways that you can get more juice. You can add more apples, or you can get a more efficient juicer that will get more juice out of each apple.

For the first option, obviously, with fruits, there is a physical limit on how much juice you can possibly get out of a single apple. But I like to imagine that if you bought a crummy juicer from the clearance aisle at Walmart, it might only extract sixty percent of the juice from each apple. If you bought a nicer juicer, maybe you could get more like 97% of the juice from an apple.

The other option means you either need to get a bigger juicer, or just take a lot more time reloading the juicer with more and more apples. In this sense, you might find yourself sacrificing performance so that you can have a smaller juicer. Not everyone has the space for a 400 apple-capacity juicer in their kitchen. And sometimes you want a nice, portable juicer that you can throw in your bag to use on airplanes and in coffee shops.

This may not be how juicers work at all, but it is how computers work. If they use too much power, they get bigger and hotter, and become something you want less and less to have on your lap while editing video or making cider. If you want more performance, you can either get a bigger one, or get one that is more efficient. And of course, past a certain point, you can’t get anymore juice out of an apple, so too there seems to be an almost physical limitation on how much performance you can get out of a laptop. If you want a juicer that only juices one or two apples at a time, you’ll only ever get 100% of the juice each apple^[This is like the fourth time I’ve capitalized Apple wrongly in this post and gone back and corrected it. I should have used a different metaphor.] has to offer. With laptops, it’s sort of felt like we’ve had that same kind of limitation, and over the last 5 years or so, juicer technology has advanced from being, at best, able to extract 90% of the juice from an apple, to extracting more like 94.5%. And we’re all wondering if we’ll hit 100% and not see anymore advancement until something that sounds sci-fi and magical comes out, like “quantum computing.”

What everyone expected from Apple^[The brand this time] was that they would jump from 94.5% efficiency to, I dunno, 99%. What it feels like they’ve done, though, is make a juicer that is—if I could quote nearly every Doctor Who companion ever—bigger on the inside. Physically it is the size of an apple, but you can just keep packing apples in there. It’s the Narnian wardrobe of juicers.

You see, the Apple MacBook Air M1 uses about 15 watts of power for both it’s graphical work and it’s more number-crunchy work (the graphics processor and the general CPU). Most of the time, it doesn’t even use that. Most of the time, it seems to hang out in the 10 watt range. The computers it’s beating are the those that give 45 watts to just the general CPU—the one for the number-cruncy work. This is a juicer that looks like it holds one apple, but takes in more like five or six apples, and extracts something like 98% of of the juice. To companies like Intel watching, it feels like they just broke physics. That’s why when Apple announced these chips and their performance, no one believed them. But the benchmarks are out, and now we all know better.

This power usage means that tasks that used to take up your entire battery on last year’s fastest Intel MacBook Pro will now use only 17% of your battery on an M1 Mac, and still finish in roughly one third the time.

Now, it’s important to note that this performance per-watt likely won’t scale linearly. Meaning if instead of giving the M1 fifteen watts, we give it thirty, it won’t literally double in performance. In some cases and in some ways, it might come close—but not in all. And that’s due to the fact that certain kinds of computer operations scale much more linearly than others. But the notion that some kinds of computer tasks perform better with certain kinds of chips is another reason to be really excited about Apple Silicon. And it’s also the topic for a future post.