“With the increased core count and frequency upgrades, Apple has a good chance to overtake Intel in the PC space” — Wayne Ma, The Information
And yeah, that’s more than a little wonky, because Apple’s silicon roadmap isn’t just about core counts and frequencies, any more than Intel’s is just about goosing voltage, but before we get into what we already know about M2, and why M3 might be another major leap forward, let’s just address the elephant-sized silicon in the room.
Because The Information’s isn’t just talking about where Apple’s going next, they’re comparing and contrasting them with Intel. Including Intel CEO Pat Gelsinger’s top priority of… winning Apple’s business back. In his most fevered shareholder dreams, getting Intel’s hot and hungry cores back inside the Mac, but more realistically, winning their fab, or fabrication business away from Taiwan Semiconductor, or TSMC. Or at least part of it.
And, even there are geopolitical arguments and, personally, I’d just love to see more chips being fabbed in North America, that’s going to be an incredibly tough win, because:
One, TSMC is already on their second generation 5 nanometer process and are driving like speed-force fast towards 3 nanometers — and more on both of those in a quantum realm minute — while Intel is still trying to stick the landing on their 10-lets-re-brand-it-as-7 nanometer process.
And two, Intel seems intent on wooing Apple back through… the lamest, most ratio’d attack ads imaginable. I mean, the last one made me cringe so hard I almost pulled a hammy. So, I don’t know if they think they’re negging, which is gross, or just not thinking at all, but the best and kindest advice I can give Intel, from the obviously scorn c-levels to the stuck-in-the-90s marketing org is just… shut up and ship. Let your silicon do the talking.
Because that’s what Apple’s been doing. Hell, they’ve been letting it scream. Including, A15, which some Benchmark LARPers claimed was proof-positive Apple had hit a brick wall and was suffering from a brain-drain, or something, but which turned out to be just full of silicon surprises by the time folks like AnandTech got all up into it. All down into it?
I’ve done a whole entire explainer on A15 already, which I’ll link to in the description below the like button, and I’ve got an M1 Pro / Max explainer on the way as well, so hit subscribe so you don’t miss it.
And that’s important because M1 was based on A14 generation IP — Icestorm efficiency cores, Firestorm performance cores — not that Firestorm! — G13 graphics cores, the whole bit. But M2, if it comes out in the next… 6-9 months, M2 will almost certainly be based on A15 generation IP — Blizzard efficiency cores — not that Blizzard! — Avalanche performance cores, G14 graphics cores, and more.
Now, the Information and Wayne Ma don’t actually say much about M2, at least not compared to M3, which I’ll get to in a Pym-particle-process-shrunk second. Just that M2 will be fabbed on TSMC’s upgraded, plussed-out 5 nanometer process.
For its second-generation processors, Apple plans to manufacture them using an upgraded version of the 5-nanometer process, two people with knowledge of its plans said. One of the people said the chips will contain two dies.
Which makes the kind of sense that does, given A15 is already on the N5P node. And that there’ll be a single die version, code named Staten, in the upcoming MacBook Air redesign. And if you want to see a dedicated video on that next Air, just let me know in the comments. But that there’ll be a dual die version as well. I just covered the potential dual and quad-die versions of the M1, reportedly coming our way in the next Mac Pro, so I’ll link that below the like button as well.
Now, Bloomberg’s Mark Gurman has also reported that M2 will stick with 4 e-cores and 4-p cores architecture, but increase the GPU core count from 7 or 8 to 9 or 10. And that, like M1, it’ll also be used in Apple’s other ultra-low power machines, like the 13-inch entry-level MacBook Pro. Or MacBook Air Pro if you’re salty.
But, if M2 really is based on A15, like M1 was based on A14, then we kinda already know a lot more as well.
Where M1’s e-cores and p-cores were… a song of ice and fire… just with a way, way better ending. M2’s double up on cold. Like Mr. Freeze level cold. Not that Mr. Freeze!
The Avalanche performance cores, other than running at 8-10% higher frequencies and having double the system cache, have also increased the L2 cache size by half again. Taken together it just makes the performance cores much more efficient. 17% more efficient at peak states. At the same time, the Blizzard efficiency cores also offer much better performance, up 23% better.
Are you getting it yet?
Yes, sure, absolutely, performance efficiency of that magnitude are important for a thermal envelope the size of the iPhone 13, but it also just so happens to be important for ultra-low power fanless computers like the upcoming even-more-Air-y-er MacBook Air. Where the current version gets stupefying battery life but does ramp down frequencies on sustained heavy loads of like… 10 to 20 minutes or more. A cooler version in a cooler, lighter MacBook Air would be even cooler.
Especially if they also contain a version of the A15’s new media engines, including and especially the ProRes encode/decode engines. Apple’s already put those blocks on the M1 Pro and the M1 Max, and it’s possible Apple will keep them exclusive to the Pro and Max variants of future chipsets as well, but how cool would it be if even the Air could just render and transcode off the big compute engines. You know, just fly casual.
On my old Intel i9 MacBook Pro, hitting export on a big video would bring the CPU to its knees and trying to do anything else was an exercise in patience and frustration. On my new M1 Max MacBook Pro, hitting export isn’t just exponentially faster, it’s pretty much imperceptible as well. Like having a second MacBook Pro to use while I’m waiting for the export. And that’s workflow changing. A MacBook Air wouldn’t need Pro versions, but there are all sorts of clever mainstream use-cases Apple could find for lightweight versions of those new media engines as well. So all the silicon fingers crossed.
And, who knows, just like M1 Pro and Max have those A15-style media engines already built in, merged like Devastator… or Voltron, depending on where your franchise loyalties lie, maybe M2 Pro and Max will have some hybrid A16 features as well? What, iPhone 14 can’t have all the fun!
And just like Apple went from TSMC’s 7 nanometer process on the A13 to 5 nanometers on A14 and M1, they’re reportedly going to be going to 3 nanometers on the A16 and M3, code named Ibiza. Which, yeah, starts to sound sub-atomically small.
Wayne says it’ll debut in a future iPad Pro before going into a future MacBook Air, and then getting built out into Pro and Max models as well. Which, yeah, Apple works at least 3 years out on their silicon, so specific product launch plans can and will shift in the interim. So expect it when and in what you see it.
But, basically, a process shrink means you can fit the same amount of transistors into a smaller space, increasing efficiency, or even more transistors into the same about of space, increasing performance, or a little bit of both, increasing both. So that’ll be a huge advantage for Apple again, right out of the gate.
At the same time, Apple is also expected to adopt ARM’s next-generation instruction set architecture, or ISA — ARMv9. Some people are expecting more better big gains from that as well. And, I mean, I’d love to see it, but Apple’s largely been driving ARM since v8 and ARM64, and I don’t think that’s changed. My read is v9 looks like ARM’s way of back-porting a lot of what Apple’s already been doing over the last half-decade to standardize it and make it available to their other licensees, so aside from some matrix multiplication boosts, which assuredly aren’t nothing, I don’t think it’ll matter anywhere nearly as much to Apple. That, and I wouldn’t be surprised, like at all, if Apple announces their own ISA one day as well. SwiftISA, and the branding goes full-circle!
So, that’s M2 for the 2022 MacBook Air, and M3 for the 2023 MacBook Air, and whether Apple makes new Pro and Max versions of every M-series system-on-a-chip, or alternates like they did with A10X and A12X, but no A11X or A13X, or puts every generation into every Mac, like they’ve been doing with every iPhone, I’ll dive into in a follow up video. So, seriously, hit subscribe. But either way, any way, this type of rampant escalation in generational IP, process shrinks, core counts, and now dies, this… multi-vector attack on all the old preconceptions about what levels of power could be reached using… so little power… it’s going to be hard for Intel or anyone to match, at least for a while.