Based on the first M2 chip launched late last year, the Pro and Max versions of Apple’s SoC flex their muscles. In addition to a doubling of CPU cores, Apple has mainly operated an explosion in the number of GPU cores, bandwidth and memory. Supplements that boost performance. But which have a cost.
In the Apple Silicon family, I would like the two new chips in the M2 range: the M2 Pro and M2 Max. Two new chips which are appearing in the new Mac Mini and Macbook Pro and which have the difficult task of making Mac users want to take the plunge of machine renewal. A difficult task as the first M1 Pro and M1 Max have – rightly – good press.
Read also: Apple announces its M2 Pro and M2 Max and updates its MacBook Pro and Mac mini (Jan. 2023)
Based on the same architecture as the first M2 of the name, but different from it in their organization, these two chips are two sisters. The M2 Max being a “bloated” version of the M2 Pro, both on the memory bandwidth side (and incidentally, the maximum amount of RAM) and on the GPU side, THE area in which Apple wants to make a difference.
Apple stays at 5nm
Like the M2, the two new chips are engraved in 5 nm “second generation”. Understand that instead of the TSMC N5 node used for the M1 chips, Apple has preferred the N5P node here. An improved version of this chip burning method that offers 7% more performance at equivalent power consumption, or 15% less power consumption at the same performance level. What is important to point out is above all that the N5P node does not require any modification of the designs compared to the N5. Apple saves money by keeping the same tools and the same fundamentals.
As for the structure of the units making up the different blocks of the chip, Apple does not seem to have made any changes to the structure. Neither in the CPU, nor in the CPU. Nor even in the NPU, which still retains its 16 computing units. The main difference between the first M2 and its two big brothers therefore remains the number of CPU & GPU calculation units and the memory bandwidth. And in one case as in the other, Apple has multiplied the breads!
M2 Pro: 40 billion transistors
To the 33.7 billion transistors of its ancestor the M1 Pro, the new M2 Pro adds 20% to reach 40 billion transistors. The chip, very different from the M2, has twice as many transistors as this one and turns out to be, as we will see, the “base” of the M2 Max. We said it above, the units of calculations do not change, it is their number which is increased.
On the CPU side, the four low-power cores are retained, but the number of high-performance cores increases from four to eight, and thus doubles the L2 cache memory from 16 MB to 32 MB. But it is on the GPU side that the increase is the largest with an increase from 8 to 19 graphics cores! These increases in calculation units are accompanied by two doublings: that of the memory bandwidth from 100 GB/s to 200 GB/s for the M2 Pro, as well as that of the maximum unified memory (from 16 GB on the M2 to 32 GB on the M2 Pro).
The M2 Pro is not only deployed on Macbooks but also, for the first time for a “Pro” chip, on the Mac Mini in a version with 12 CPU cores, 19 GPU cores and 16/32 GB of unified memory. What give a big boost to these machines.
M2 Max: 67 billion transistors (which mainly benefit the GPU)
Take an M2 Pro, “stretch” the GPU area and “voila”, you have an M2 Max. Or almost, since it was also necessary to work on the memory bandwidth. On the CPU side, therefore, nothing differs between the two chips: four low-performance cores and eight high-performance cores – by the way, Apple has not communicated on possible frequency differences.
But the GPU benefits from a doubling of calculation units, which go from 19 to 38 units! It is thanks to this super graphics processor that Apple intends to seduce the creative professions. Editors, animators, modelers, etc. All professional applications using GPU-accelerated filters will benefit from this doubling of power. Which promises, according to Apple’s marketing video ” not even realizing that we are only on a laptop “. Still happy, you might say, since those who need more juice will still have to wait a while to see a replacement for the M1 Ultra arrive.
This deluge of graphics power has a great need for memory, both in quantity (we go from 32 GB to 96 GB maximum!) and in bandwidth. Which is again doubled: we go from 200 GB / s on the M2 Pro to 400 GB / s on this M2 Max. The disappointment for our technophile and gamer friends being that Apple does not communicate about improved compatibility of its drivers with games. Moreover, in several areas, Apple does not communicate. Or as he sees fit.
No comparison with the current PC world
When it was launched last November 2020, the M1 kicked things up a notch. In addition to the traditional forces of the verticality of the control of Apple which controls at the same time the operating system, the drivers and the chip, the American could count on the rise in power of the hearts ARM. As well as on a 5 nm engraving which gave it an advantage over the Intel chips of the time (10 nm in the best case, 14 nm most often) and from AMD which had just accessed 7 nm.
Read also: MacBook Pro 13-inch M2 review: Apple changes nothing but the essentials (June 2022)
But this time, Apple’s comparisons are only for older products. In its presentation video, the technology giant only talks about its old terminals, whether it is the generation in M1 (N-1) or products formerly equipped with Intel chips (N-2). Apart from the fact that the comparisons are not always very honest – the GPUs integrated into the processors chosen by Apple were not the most efficient in the Intel range at the time – we especially regret that Apple does not want to compare itself to the PC world.
Read also: Will the Ryzen Mobile 7040 be the essential laptop processors of 2023? (January 2023)
We are thinking in particular of a few chips, such as the future Ryzen 9 7040 called “Phoenix” from AMD. Three 35-45W mobile SoCs equipped with 12 GPU cores from AMD’s latest generation (RDNA 3). With its only 25 billion transistors (the die of the chip measures only 10×17 mm), this chip seems to have what it takes to measure up to the M2 Pro – and to tell the truth, one legitimately wonders what the performance of this chip would be with a doubled GPU like at Apple?
Finally, when we look at the promises of generational performance gains, we wonder if Apple hasn’t already rinsed its architectures.
Transistor density and chip prices
Before the mass of Apple fans start claiming the superiority of the Californian brand in chip design, two things must be taken into account. On the one hand, these performance increases are, after all, quite mechanical. If Apple promises, at equal range, GPU improvements of up to 30%, the CPU is at 20%. Which is almost exactly equivalent to increasing the number of transistors. In short: if Apple has nevertheless found tricks to gain a few small percentages of performance here and there, it is above all the number of transistors in addition that make the difference.
Then comes the reminder of the cost of these chips: they are disproportionate to what the PC world offers. Because Apple’s promises in terms of performance gains are always based on the most efficient chip in the range. Between a MacBook Pro 16 M2 Pro 12 CPU/16 GPU displayed at €2629 (with a 1TB SSD) and its version equipped with an M2 Max 12 CPU/38 GPU at €4849, there is a difference of €2220. A huge supplement over the PC world, where an RTX3080 Mobile will arguably do just as well when it comes to professional acceleration. But will largely dominate in games.
The fact remains that Apple still scores a major point in the world of SoC CPUs: that of the performance-per-watt ratio and the support for AI calculations by its NPU. An unchallenged domination that once again makes us wonder about the lack of reaction from Intel and AMD. Who now both seem to have the cards in hand to develop a SoC as complex and integrated as Apple’s chips.
