M1 Pro Max vs M2 — Not What You Think!

In this video, I’m going to explain the exact difference between Apple’s M1 Pro and Max, and M2. Between ultra high performance and ultra low power, massively more multi-cores and next-generation, better cores. Because we’ve got a slew… a slew of new Macs heading our way this year, starting as soon as next month, and M2 just has to be better than 1, right? Well, no, wrong… kinda.

Hit that subscribe button and bell to help us build the best community in tech, and then let’s dive super nerdy deep.

Come the holidays, we may well have an M1 iMac, M1 Pro and Max iMac, and M2 iMac to choose from. Same with the Mac mini and MacBook Pro. Entry level, powerhouse, and premium all. And that might just seem confusing and… yeah, convoluted as hell. So let me break out ye old Apple quadrants… decrants… tesseracts… whatever, relax about it.

When Tim Cook first announced Apple was transitioning from Intel to custom silicon, senior Vice President of hardware technologies, Jony Srouji said that would include a family of SoC, of systems-on-a-chip.

So, the M1 family actually began with the A14 Bionic, the chipset in the iPhone 12 that debuted in October of 2020. With 2 firestorm performance, or p-cores at up to 3GHz, 4 icestorm efficiency, or e-cores at up to 1.8GHz, and 4 G13 graphics cores, or GPU. Also 16 neural engine cores, or ANE, H.264 and 265 encode decode engines, and up to 6 GB of unified memory at 42Gbps.
That all translates to up to 1583 single and 4210 multi on the geeky benches, at least in the relatively roomy iPad Air.

Now, one of the ways you can scale a chipset architecture is by escalating the number of cores. That’s what Intel has done for years with Core i5 vs. Core i7 vs. Core i9, for example. Apple as well, with A12, A12X, and A12Z. Same generation of cores, just more and more of them.

The scaling actually goes both ways; there are two A13 e-cores in the heart of the Apple Watch system-in-package, which is key to Apple’s whole scalable system architecture advantage, but… topic for another video.

With A14, instead of making A14X, Apple made M1 instead. 4 Firestorm p-cores at up to 3.2GHz, 4 Icestorm e-cores at up to 2GHz, up to 8 G13 graphics cores, up to 16 GB of unified memory at 70 GB/s, and 2 each USB and Thunderbolt controllers. In other words, the exact same cores as A14, just 2 more performance and up to 4 more graphics, tuned a little faster to take advantage of the bigger enclosures.

Or, up to 1718 single and 7426 multi on the marks.

But, when you normalize for things like frequency and thermal envelop, including active cooling systems, and the ability to run faster, hotter, longer inside machines that are bigger than a phone, you get approximately the same single core performance, but way more multicore performance, because… just way more cores.

And, yeah, turns out when you run that cool, you can also run faster, longer, and get way better performance, especially per watt of power, in all of those Macs.

Which gave Apple room to keep escalating the core counts as well.

Enter M1 Pro and Max. Still built on that same A14 Bionic generation silicon IP. But with only 2 e-cores at the same 2GHz, because the focus was less on low-power tasks like checking email and more on high-power tasks like compiling code. Which is why they offer up to 8 p-cores as well, also at the same up to 3.2GHz, but in dual clusters. I’ll link up my deep dive below the like button if you’re curious about all that. Also, up to 32 GPU cores. All fed by up to 64 GB of unified memory, up to 400 Gbps of memory bandwidth, up to 2 ProRes media engines, and 3 USB and 3 Thunderbolt controllers. In other words, the exact same cores as M1, but up to 4 more p-cores, 28 more GPU cores, those ProRes engines, and… ungodly amounts of RAM.

That’s up to 1747 on single, but 12239 on multi.

Again, way more massively multicore performance, because… way more massively multicores. And single core performance ain’t nothing, not when you’re counting dock bounces for an app launch, or enjoying that instant iPad-like responsiveness on the Mac, but these days almost all workloads are almost all multicore aware, so getting all those multiple cores really, really pays off. But, of course, as each core improves, so does the sum of all the multiple cores.

Which brings us to the other way you can scale a chipset architecture, by improving the next generation of cores. That’s what Intel has done for years, originally with their famous tick tock cycle, more recently with their infamous tick, tock, optimize, optimize, optimize cycle, but like Ice Lake, Tiger Lake, and Alder Lake, for example. And also Apple as well, most recently with A13, A14, and A15 Bionic.

Some years, there’s a process shrink, like going from 7 nanometers to 5 nanometers, meaning Apple can fit more transistors in the same amount of space, or the same amount of transistors in even less space, which translates into more performance at the same power or the same performance at less power, or, more often then not, the best balance they can of more performance and less power. Other years, there are architectural improvements like bigger system cache, which improves speed and efficiency by reducing callouts to slower, hungrier main memory.

A14 was the former. A15, the latter. And if we see M2 any time before the end of 2022, the iPhone 13’s A15 Bionic is almost certainly what it’ll be based on, just like M1 was based on A14.

A15 has 2 Avalanche p-cores at up to 3.2GHz, 4 blizzard e-cores at up to 2.0GHz, and up to 5 G14 GPU cores. 16 next-generation ANE cores, ProRes included alongside H.264 and 265 in the media engines, 6GB of unified memory at 42GBps.

That’s up to 1690 for single, and 4645 for multi.

But also where you start to hit into the limitations of benchmark LARP. Because it’s increasingly hard to see what’s hitting an e-core vs. a p-core, or a media engine vs. a GPU. And that’s particularly important for A15, where the e-cores got way more performance, the p-cores got way more efficient, and those media engines didn’t necessarily take work off the CPU like they did on the Mac, but enabled work that wasn’t even possible on a phone before. So, faster, cooler, and yeah, just plain cooler. And I’ll link that deep dive below the like button for more as well.

So, if we choose to live dangerously and assume past behavior is still the best indicator of future behavior, meaning Apple continues to do with M2 what they did with M1 and A12X, we should be seeing:

4 Avalanche p-cores, maybe up to 3.4GHz, 4 blizzard e-cores, maybe up to 2.2Ghz, and up to 10, count ‘em 10 G14 GPU cores. That ProRes engine. Extra system cache, but probably still up to 16 GB of memory, and 70Gb/ps bandwidth and just 2 USB and 2 Thunderbolt controllers.

So, not much more capability in M2 over M1, but more performance at the same efficiency, even and including a little taste of that ProRes power. Compared to M1 Pro and Max, better single core, the way A15 is better than A14 at single core, but not at all competitive on multicore, not even close, because it simply doesn’t have enough of those multiple cores to compete. Where it should win, though, is battery life — like iPhone 12 to iPhone 13 win on battery life. Everything else being equal.

Now, if we don’t see M2 any time before the end of 2022 or the beginning of 2023, then it’s possible it won’t be based on A15 at all, but the next generation A16 coming with the iPhone 14.

Unlike A15, A16 is expected to benefit from a process shrink, going from 5 nanometer to 4 or 3 nanometer. Those are marketing names, not consistent measures, but it just means even more efficiency and performance, depending on how exactly Apple spends and saves the new transistor budget it’ll enable.

And that would, legit, translate to even better single core performance, even better battery life, but still nowhere nearly the massively multicore performance, because still nowhere nearly the massive amount of multicores.

So, come the holidays, and having the choice between M1, M1 Pro and Max, and maybe even M2, here’s how it’ll break down. M1 will be not only the ultra-low-power option, but the ultra-affordable entry level, holding the line on up-front prices. So, if you just can’t or won’t spend dime one more than you absolutely have to on a new Mac, M1 is still going to be your best friend.

M1 Pro and Max will be still be top of Mac mountain when it comes to total performance potential though, especially if reports of dual and quad-die M1 Max variants in the new iMac Pro or Mac Pro prove accurate. Expensive, yes, but meant for those who value time even more than money. And nothing will come close to that until there’s an M2 Pro and Max or M3 Pro and Max, depending on when and how Apple chooses to update that ultra high performance silicon. And it took almost a year to go from M1 to M1 Pro and Max anyway. So, if you really need Pro… even Max level performance, you’re still going to want M1 Pro or Max.

And leaves M2 as the new ultra premium option. More powerful than M1 but also way more efficient than pretty much anything, so Apple can slap it into the hot, new, ultra-sleek, ultra-sexy next generation Mac designs, at a slightly higher price, at least at launch. So, if you’re more than willing to pay a slight premium for maximum style meets minimum profile, you’re going to want M2.

Because that’s how scalable system architecture works, and to hear Apple say… just all of that out loud, check out my interview with their VP of silicon and VP of Mac Product marketing.