Technically, the M2 Ultra question is less a question of “if” than a question of “when”. As for its theoretical performance, the excellent architecture of Apple already allows us to draw the outlines. It remains to be seen, however, whether the market needs such a chip. And to ask the question of the future of Apple chips.
Center thirty-four billion transistors: here is a chip that does not yet officially exist, but whose (very serious) hypothesis is enough to make you dizzy. This chip is obviously Apple’s M2 Ultra, expected to succeed the M1 Ultra. With the arrival of the M2 last year and the high-performance versions of the M2 Pro and M2 Max yesterday, all eyes are on Apple for an “Ultra” version.
Read also: M2 Pro and M2 Max: the new Apple processors under the magnifying glass (January 2023)
A chip that would replace the M1 Ultra, which was already the superlative processor when it was launched in early 2022. Integrated into the Mac Studio, it is still the most complex and sophisticated all-in-one (SoC) chip to date. rich in mainstream market transistors. Or rather a double SoC, because it is in fact two M1 Max chips (57 billion transistors) “stuck” together to offer 114 billion transistors, which is more than many data center chips! While the M2 Pro and Max put a lot of emphasis on the GPU, the M2 Ultra would be, like its ancestor, a pure and simple doubling of the calculation units of the Max.
Why the M2 Ultra is credible
The design of the M2, M2 Pro and M2 Max is in line with the previous family. The basic M is a low-power chip that has its place as much in an ultraportable as in an iPad. The M Pro is a new chip that lays the foundation for an M2 Max to which we double the GPU units. In doing so, one only has to look in the rear view mirror and see the M1 Max’s interconnect units to imagine a M2 Max so equipped to ‘mate’ with another M2 Max to form an M2 Ultra.
Read also: Apple announces its M2 Pro and M2 Max and updates its MacBook Pro and Mac mini (January 2023)
Apple having remained in 5 nm and having kept the organizational scheme of its computing units, no hardware brake opposes it. No more than software, since the two bricks that are the CPU or the GPU can easily benefit (depending on the applications) from a doubling of the calculation units. The only variable to take into account is that of the market. Namely, is the audience, already quite small, of current users of the M1 Ultra large enough to justify the existence of an M2 Ultra so quickly?
The promises of an M2 Ultra
No need to be a guesser to predict the theoretical performance of this chip: “Up to +20% in CPU power and up to 30% in GPU power”. Where do these numbers come from? Simply presentations of the M2 Pro compared to the M1 Pro and M2 Max compared to the M1 Max. The M1 family having proven the “scalibility” that is to say the ability to increase in power constantly. Apple has achieved a tour de force not only in the development of its chips, but also in its framework and its software drivers, all recent programs compiled in ARM benefit almost linearly from the increase in the number of cores as well as from a possible increase in frequency – for the time being, THE unanswered question at the presentations of the M2 Pro and Max.
At constant or increasing frequencies, and taking into account the selected node (N5P a bit improved compared to the N5), we can also be almost sure that both the M2 pro and Max and the potential M2 Ultra consume more energy than their predecessors. Because as we can see on the plans of the “ die (pronounced “daie”, the silicon heart of the chip) communicated by Apple, the processors are physically bigger, because they embed more transistors without any real reduction in size. Apple’s architecture being the queen of the performance/watt ratio, this is not a huge concern, but there is still a small risk of throttling. A phenomenon where frequencies (and with it performance) automatically lower to reduce heat dissipation when the system gets too hot. However, a doubling of the chip implies a doubling of the additional heat released: could the Mac Studio get by without worry? To this question, only the thermal engineers of Apple have the answer!
Architectural limits in sight?
To compose an M1 Ultra, Apple glues two M1 Max chips that it interconnects. For this, the need for bandwidth is important and it seems to us quite credible that the technical blocks designed by Apple are able to withstand the +20%/+30% performance offered by the new chip. But before even talking about the M3 generation, we must point out the perceptible limits of Apple’s current CPU architecture. By using a slightly better node, Apple makes its chips 20% bigger to get 20% more CPU performance. Proof that a good part of the performance gains between the M1 and the M2 is due to the increase in the number of transistors – on the GPU part, on the other hand, Apple seems to have made further significant progress.
Read also: Apple Mac Pro: why the M2 Extreme should not see the light of day (Dec. 2022)
And it seems that Apple is stuck, one way or another, for the design of an M2 Extreme grouping four M2 Max. Whether for cost reasons (TSMC yields, market demand) or for interconnect limit reasons. In any case, many analysts assure that the Mac Pro are still being pushed back. Because the potential M2 Ultra chip would not be at the level of Apple’s requirements to ensure the succession of current machines, powered by Intel Xeon.
And that’s also where you have to look: if Apple has succeeded in developing chips with the best performance/watt yield, the Intel, AMD and Nvidia clique still has a large advantage in pure power. This, they owe it to their architectures, sometimes less efficient in the lower consumption brackets (even if this is tending to change, thanks to the influence of Apple), but capable of anything when you “make the watts explode”.
Max and Ultra: unique chips in more ways than one
Apple has built itself a unique playground with its Max and Ultra chips. Unique SoCs by their number of transistors, by their memory management… and by their cost. At 114 billion transistors, the M1 Ultra and 134 billion for the hypothetical M2 Ultra, these chips greatly exceed a CPU + GPU couple separated from other manufacturers such as Intel, Nvidia and AMD. Additionally, the implementation of memory modules soldered within millimeters of the chip provides a “unified memory” pool unique in the world of consumer computing. A design that only Apple can offer, the other three players being platform suppliers to manufacturers. Who, themselves, need to create ranges, but without mastering what happens in the chips.
Finally, there is the question of the price of the chips. Between a Mac Mini and a Mac Studio, the chassis cost differential is quite marginal. The real difference is in the processor: between an M1 (and an M2 from the end of the month) from Mac Mini and an M1 Ultra from Mac Studio, it’s almost €5,000 difference. A price difference justified by the chassis, therefore, the amount of memory grafted onto the chip. But also and above all at the cost of a giant double chip whose yields are infinitely lower than those of a simple M1/M2.
In any case, Apple has developed a category of chip that the competition cannot or does not want to produce. Allowing it to maintain a key differentiating advantage. Let’s hope, however, that the company does not suffer too much from the brain drain of a great many brains from its semiconductor division. Because the competition is not sleeping: Intel will accelerate the deployment of its advanced nodes, flex its muscles in the graphics and design more and more efficient cores. AMD already has the technical cards to design an equivalent chip (but the market is not there yet?). And Nvidia remains (by far!), the king of pure power AND graphics efficiency. Apple started a revolution and showed the way, now it has to remember that it is not alone!
