If we were to summarize Apple’s M2 chip in a few words, it’s writing that this is a small development of the M1 chip that the company introduced in 2020. So no revolution on the program, but an upgrade in small steps to bring a bit more performance, while further limiting the energy footprint. The M2 chip as presented is therefore not intended to completely bury the M1 generation, the Pro, Max and Ultra versions remain much faster.
If we refer to the “picture” of the chip presented by Apple, the M2 is significantly larger than its predecessor. A point that is not surprising when we look at the number of transistors, up 4 billion to reach a total of 20 billion. Two solutions were then offered to Apple to accommodate these additional transistors: change the fineness of engraving to preserve a similar surface (or even reduce it), or enlarge the surface of the chip while maintaining an engraving process identical to the one used on the M1. The company chose the second option by again using a 5 nm process, however a little more advanced (TSMC N5P without a doubt), which guarantees a small optimization of efficiency.
At first glance, the processor part does not appear to move a toe part. There is a hybrid configuration with four high-performance cores and four low-power cores. Unfortunately, Apple provides only a few technical details about these cores and is careful not to define their exact nature.
But by proclaiming an increase in performance of at best 18%, the company is giving us some clues. The high-performance cores should therefore be based on the Avalanche architecture of the A15 Bionic processor on the iPhone 13 and iPad mini 2021. No revolution compared to the Firestorm cores in the M1, but some small optimizations and an ability to keep a higher operating frequency high.
The biggest change is without a doubt to be found on the side of the low-power cores, which should be based on the Blizzard architecture of the A15 Bionic. If they are still aiming for energy efficiency, they are much faster than the low-power Icestorm cores in M1. Which greatly increases the performance of applications that jointly utilize the chip’s eight cores. These small architectural improvements – because let’s remember, these are clearly more optimizations than radical changes in architecture – are linked to the classic trick of processor designers, namely the use of a larger cache. The 12 MB L2 cache is thus given up in favor of a 16 MB L2.
These improvements are likely to be combined with an increase in operating frequencies, which in turn allows for performance improvements. Apple has also given a boost to its memory interface, which now peaks at 100 GB / s, against 68 GB / s on the M1. This increase in bandwidth is approved by a switch to LPDDR5-6400, still of the unified type. Beneath this term lies the idea that memory can be used by the system as well as by the graphical circuit. It is also directly connected to the M2 chip by being placed on the same physical envelope.
Apple apparently did not just make a few changes to the processor part. The graphical circuit has also been reviewed, although we are not aware of the exact nature of the improvements made. For example, Apple is talking about a Next Generation architecture, much faster with a raw power of 3.6 TFLOPS, versus 2.6 TFLOPS for the GPU on the M1. Although adjustments and optimizations have been developed for the GPU and its operating frequency, this boost in performance is primarily due to the use of 10 cores instead of eight cores, resulting in a small increase in energy footprint (15 W versus 12 W). A 25% increase in the number of cores, resulting in a 35% increase in performance. This last “surplus” is not explained, but we note that Apple mentions an enlarged cache, which combined with a frequency, which in turn is undoubtedly revised upwards, leads to the said 35%.
The block dedicated to artificial intelligence is also under development. Neural Engine is a 16-core type and probably also taken from A15 Bionic. It is capable of processing up to 15.8 trillion operations per second and is 40% faster than the M1’s neural motor. Media Engine, a block dedicated to video management, is evolving steadily. It still allows for faster encoding and decoding of H.264 and HEVC video streams, but is no longer limited to 4K streams and is now capable of processing 8K streams. It is also capable of speeding up the processing of Apple ProRes streams, but still ignores the AV1 format, which is gaining momentum nonetheless. Finally, Apple has not found itself forced to implement support for Thunderbolt 4, thus always limiting the administration to Thunderbolt 3. USB4, on the other hand, is good on the program, with a maximum of two ports controlled by the chip.
There is still the question of the behavior of the M2 chip towards the competitors. If Apple was careful not to stick to the comparison exercise a few years ago, the company now delivers some sparse clues through some imprecise curves and poor details. For processor performance, the Apple M2 places it against the Intel Core i7-1260P, a mid-range chip targeted at ultrabooks. The processor in the M2 is advertised as slower in absolute numbers (difference of about 13%), but much less greedy in terms of power consumption, the M2 only requires about fifteen watts to operate at 35 to 55 W for the Intel chip. Which allows a much quieter operation – the MacBook Air M2 is also completely passive – and achieves significantly greater autonomy, which was already the case with the MacBook M1. Apple also mentions superior performance over the Core i7-1255U, but is not comparable to AMD’s Ryzen solutions.
In terms of graphics performance, of course, Apple did not want to offer too much comparison, and contented itself with developing performance more than twice as high as on the Intel Core i7-1255U iGPU. Still, we are talking about an integrated graphics circuit with a memory bandwidth of almost 100 GB / s. In all likelihood, and in particular based on the announced performance increase of 35% compared to the M1, we should not see a revolution.
We are betting on a solution that is capable of doing better than Intel’s integrated graphics circuitry and comes pretty close to what AMD is offering on their latest Ryzen 6000 mobiles based on RDNA2 iGPUs. Not enough to turn the MacBook M2 into a slot machine, though the MetalFX Upscaling technology announced at the same time as the M2 will undoubtedly allow for beautiful things. For our part, we are quite eager to see how the hardware acceleration in the iGPU behaves on compatible applications, for it is much more in this field that Apple has a card to play.
We can’t wait to test the MacBook Air M2 and MacBook Pro M2 in our labs, the first two laptops to use Apple’s new chip. The opportunity for us to confront this promising SoC against the new processors from Intel and AMD. However, it will be necessary to be patient: pre-orders are not yet open and delivery of the first models will not take place until the current month of July.