How the Mac’s M1 chip gave it a second life

Given the choice, Apple would have rolled out its first custom-designed Mac chips on its own terms, probably at a high-profile event in the Steve Jobs Theater at Apple Park. But given that it was November 2020, the company was forced to release a 45-minute video instead.

No amount of in-person theatrics would have upstaged the star of that show, the M1 processor. Five years later, it’s clear that the arrival of Apple silicon has utterly changed the trajectory of the Mac.

A careful start

In that first event (which you can relive in the YouTube video below), Apple announced its first wave of M1 Macs: the MacBook Air, 13-inch MacBook Pro, and Mac mini. The Macs themselves all used the same design as their Intel predecessors, as Apple wrapped potentially scary new technology in completely familiar shapes.

Then the results of the first M1 speed tests arrived, and nothing felt scary anymore. Everything was fast, much faster than Intel, so much faster that even software compiled for Intel running in a code-translation layer via Rosetta ran just fine. In fact, the M1 was such a fast chip that, five years later, Apple’s still selling the M1 MacBook Air. (For $599, at Walmart.) And it’s still a pretty nice computer!

Apple’s next trick was rolling out new versions of (almost) every Mac model, redesigned for Apple silicon, as well as an entirely new model, the Mac Studio. The new chips, new designs, and a pandemic-fueled increase in people working from home all sent Mac sales soaring.

The five years before the arrival of Apple silicon were the five best years in the history of Mac sales to that point, averaging $25.5 billion a year. It was a pretty scary move to pull the rug out from under the Intel Mac era, but Apple’s move was vindicated: The first five years of Apple silicon are now the five best years in the Mac’s history. Mac sales were up nearly one-third compared to the previous five-year period, to $33.7 billion a year on average.

So it went pretty well, especially considering the huge question that hovered over Apple’s entire plan to switch to its own processors: could a chip designed for a phone ever possibly power a Mac?!

Pro considerations

It’s been five years since the Apple silicon era arrived for the Mac, but of course, Apple had been using its own chip designs for a decade before that in the iPhone and iPad. During that decade, Apple’s chip team made a lot of decisions that made sense for mobile devices, including optimizing for power efficiency and building a tightly integrated shared-memory system into the chips themselves, which itself necessitated also integrating graphics processing into the main chip. How would that scale to the exacting needs of pro-level Mac users?

Apple’s first step was actually to build a better processor for the iPad Pro. Starting with the A8X in 2014, Apple built five generations of iPhone chip variants with extra processor cores and other features that weren’t necessary on an iPhone, but might be required on a high-end iPad. The final one of those chips, early 2020’s A12Z, was not coincidentally the chip put in a Mac mini case and provided to developers in the summer of 2020 as an Apple silicon test environment.

Then, after the M1, which perfectly powered an array of Apple’s lower-end Macs, Apple had another shoe to drop: the company added Max and Pro chip designs, which dramatically increased the number of processor cores and available memory. And, it turns out, that if you take a phone-forward chip design and load it up with CPU and GPU cores and speedy integrated memory, it will actually work pretty well in higher-end Macs. (The downside: user-installable RAM and outboard graphics cards are no longer an option.)

Varying the playbook

After the M1 generation, a lot of us keen observers of Apple figured the company’s chip roll-outs would follow that same pattern. But it hasn’t been so simple. Some chip levels come and go, sometimes all the chips are announced at once, while other times they’re rolled out in batches.

On the engineering front, things are also different from generation to generation. One year, Apple might upgrade the CPU or GPU cores, or throw in a new Neural Engine. The M1 Pro and Max designs felt of a kind, but in later generations the Max came into its own while the Pro became more like a mid-tier upgrade to the base-model chips.

Maybe it’s not surprising that Apple keeps mixing it up. That decade of iPhone chip design was a great base, but certainly in five years, it’s learned a lot about what needs to be done to satisfy Mac users. Apple has even figured out how to load high-end chips up with enough RAM to satisfy even hungry AI processing loads, despite its choices to keep that memory as tightly integrated into the chip design as possible.

The result of all of this is, though every generation has its quirks, Apple has managed to not drop the ball after the gigantic leap from Intel to M1. Every generation of M-series processors has offered impressive speed boosts. Apple’s CPU cores just get 10% to 30% faster every generation. The GPU cores got faster in all but one generation–and in that generation, overall graphics performance still got faster because the chips all had more GPU cores.

Through performance improvements and by increasing the number of processor cores, Apple has managed to relentlessly improve the speed of Mac processors to the point where the just-released base-model M5 processor is roughly twice as fast across the board as the first M1.

Rectifying an error

If there’s one aspect of the future of computing that Apple’s chip designers missed, it’s AI. But even that’s not really accurate, as Apple’s been shipping what is now generally known as an NPU, or Neural Processing Unit, for more than a decade–it’s called the Neural Engine.

Apple’s miss was its assumption that the Neural Engine would be enough to handle machine-learning tasks. And while the Neural Engine definitely gets a workout and keeps getting improved, the unexpected rush to AI in the tech industry also meant that a lot of AI software was written to use CPU and (especially) GPU cores rather than a dedicated NPU.

Chip design can take a long time, but Apple reacted to the trends in AI pretty quickly. The M4 chip design added “neural accelerators” to improve AI jobs happening on its own CPU cores, and the new M5 design adds those accelerators to GPU cores. Apple’s chip designers knew AI processing would be a thing, but they needed to adapt to where that thing was actually happening–and they did.

The victim of Apple silicon

There’s one victim of Apple silicon: the Mac Pro. While Apple still ships a Mac Pro, its performance matched the performance of a Mac Studio with the same Ultra-class processor. The Mac Studio recently got an upgrade to the M3 Ultra, but the Mac Pro remains… a product in Apple’s line-up.

While most Macs have evolved to be perfect fits for Apple’s chip strategy, the Mac Pro–a traditional PC tower–feels very much like a device from a previous age. It’s unclear if there will ever be another update, or why anyone would choose it over a Mac Studio equipped with a similar Ultra-level processor.

Foundry

Never say never. If the last five years have taught us anything, it’s that there’s usually another Apple silicon plot twist waiting right around the corner.