It doesn’t take too much searching of YouTube right now to find people hyperventilating over Apple’s new baby, the ARM-based M1 CPU which represents the vanguard of Apple making another major switch in its choice of processors. Some of these videos depict the M1 MacBook outpacing a similar-looking laptop by factors approaching two to one. Can Intel really be so comprehensively outdone?
As you’ll note, we’ve been careful not to give any specific YouTube auteur the oxygen of publicity because, not to spoil the party atmosphere, some of the conclusions people are reaching wilt under serious scrutiny. We’re about to talk some numbers, and those numbers aren’t very sexy, but unfortunately “Mid-2020 MacBook Pro 13-inch” is not a spec that actually tells us much about the underlying machine. For instance, there’s people out there comparing the new M1, which is an eight-core CPU with 16MB of cache, with a 2019 model MacBook with a Core i5-8257U, which is a four core CPU with 6MB of cache.
That’s not even the current spec, but on the basis of this test we are invited to believe that the new M1 processor represents a sea change in the price-performance ratio of Mac laptops. It would be equally reasonable to conclude Apple has long been charging rather ambitious prices for laptops containing less than the highest-end technology. To put this in perspective, the current MacBook Pro 13” with the Core i5-1030NG7 (which is two generations newer than the one used in the comparison) sells for almost exactly the same price as Dell’s equally well-made XPS 13, which boasts a considerably beefier Core i7-10510U processor and otherwise similar spec. Draw your own conclusions about the overall value proposition and whether that makes the new MacBook look brilliant, or just the old one look expensive, or perhaps some combination of the two.
Still, people who buy Apple products generally know that they are paying a premium price for a luxury experience, and there’s nothing wrong with that. MacOS is widely held to be a well-engineered thing, and Apple’s hardware has long been beautifully made and oozing taste and precision, with only a few exceptions – small, black, cylindrical exceptions. Even so, with comments like “These M1s are smoking the Intels” (exclamation point, exclamation point, smiley face, smiley face) it seemed worth looking into the details. These M1s are indeed beating, if not quite smoking, at least some of the Intels, though it’s not quite clear whether they’d smoke the AMDs so readily.
The first time the Mac changed CPUs, moving away from the Motorola 68000 series that launched the original Macintosh in 1984, involved the PowerPC 603. This had something in common with the ARM cores in the M1, because PowerPCs were RISC devices, for reduced instruction set computing. Any CPU is capable of basic mathematical operations, storing and retrieving data, and other things; those are instructions and the full list of instructions a CPU can perform is its instruction set. The idea of RISC is to provide a small instruction set which keeps the CPU simple. The alternative, represented by Intel and AMD’s current CPUs, is a larger instruction set with more advanced instructions, which do more.
The promise of RISC, since its inception in the 80s, had been that a simpler CPU could be made to work faster and with lower power consumption. A huge amount of work was done on it, culminating in things like PowerPC and the MIPS R-series used by Silicon Graphics’ monster workstations. Possibly with all this in mind, Apple actually collaborated with (what was then still) Acorn on development of what would become the ARM610 used in the Newton PDA. The 610 was then used in the Acorn RiscPC, then licensed to DEC to become StrongARM and then to Intel becoming the i960. We follow this path only because Intel then developed XScale and sold it to Marvell, a name that will pop up again later. RISC seemed good to go.
But it was not to be. Did the promised benefits play out? The answer was a resounding “eh, kind of, a bit,” perhaps due to the fairly obvious caveat that with each instruction doing less, achieving any particular result needs more instructions. The fact that workstations of every stripe either disappeared or eventually went to Intel CPUs is a result both of that equivocal result, and of the huge amount of R&D poured in by Intel and its competitors. Another result of that is that most CPUs now use RISC cores, with non-RISC code being suitably translated on the fly. In the end, RISC, for all its benefits, couldn’t really keep up with the white heat of development until the explosion of phones and tablets made the power benefits particularly valuable.
So, should we expect huge real-world, like-for-like performance gains from Apple based on its move to M1? Again, eh, kind of, a bit. Reasonable tests show a gap versus Intel, which also seems to be the case when compared with AMD’s Zen 3 Ryzens, although all of that is predicated on whether you’re interested in price-performance ratio or performance per watt. Regardless, that puts paid to any remaining notion that there’s some sort of alien technology in M1.
Possibly the most reasonable hope is for noticeably reduced power consumption, which is why an 8-core chip is practical on a laptop in the first place. That might lead to greater performance through reduced thermal throttling, let alone any other benefit. Some more of the improvement might be down to the very small 5nm minimum feature size. There’s also benefit in having the memory tightly integrated onto the same device as the CPU itself, though at the cost of making it effectively impossible to create, say, a 32GB machine around this particular device, at least with symmetrical behaviour across all of its memory.
Perhaps the biggest surprise is that AMD and Intel haven’t done the same thing already. Intel and AMD have made little noise about asymmetric CPUs for laptops, with big cores for the heavy lifting and smaller ones for when work is slack. Apple’s move is also not quite as new as it might seem, with companies including Ampere, Marvell (there they are) and Amazon already throwing resources at ARM chips for server farms. The modern world of cloud computing relies heavily on warehouses full of servers – although in fact, it’s often the case that there’s just as much floor space given over to air conditioning as computers, given the heat generated. Anyone running a facility like that is extremely interested in any option to reduce the electricity and air plant costs. The fact that it hasn’t happened is likely due to both industry inertia and the fact that there really hasn’t been an ARM CPU to equal the likes of a really big Xeon.
Could it be done? Apple have probably already done it, pursuant to an ARM Mac Pro, but so have other people. Some of the biggest current ARM server chips – which we might assume would allow us to build true workstation-class machines of Xeon performance – build their horsepower through having a lot of cores, which isn’t necessarily the best approach from the software engineer’s point of view. The likes of Intel and AMD are not likely to sit idly by while the world pivots, either. Is it possible that the whole world of computing, from phones to workstations, will slide ARMwards? Sure. Is it possible that there will be a fightback, and that Apple will find itself sliding toward another incumbent in a few years? It wouldn’t be the first time.
Either way, while there’s a choice, let’s make that choice based on specifications, not advertising.
Editor’s Note: The comparisons made in this article refer to the one-up-from-entry-level Apple MacBook Pro 13.3”, Apple’s part number MVH22B/A, and the Dell XPS 13 7390, which were advertised at £1139 on johnlewis.com and £1149 on currys.co.uk respectively, at the time of writing. There are some quite noticeable regional variations in the prices of both.