How Apple builds M2 chip on M1 and sets up a strong roadmap
The M1 is a great chip. Originally an “X” variant of the A14 chip, it takes the iPhone and iPad processors and doubles the high-performance CPU core, GPU core and memory bandwidth. The M1 chip is so good that it’s equally amazing for tablets and thin-and-light laptops, as it easily outperforms any competing chip with a similar power draw for desktops, and offers similar performance for processors that use at least twice as much power.
Now, a year and a half later, and after offering three more powerful variants of the M1 (M1 Pro, M1 Max, and M1 Ultra), it’s time for the next generation. Announced at WWDC and first introduced on the new MacBook Air and 13-inch MacBook Pro, the M2 is basically a system-on-chip that we predicted it would be: the M1 key next to the A14, the M2 next to the A15. What? It is made up of 20 billion transistors, 25 percent more than the M1, and it is still built using a 5nm manufacturing process, a new advanced “second-generation” 5nm process.
Here are the most significant ways to improve M2 compared to M1.
M2 vs M1: CPU
The M2 doubles the high-performance CPU core from the A15 for an eight-core setup of four high-performance cores and four high-efficiency cores. In this case, it’s like M1.
However, there are some architectural improvements. On the performance core, the shared L2 cache has gone from 12MB to 16MB and the clock speed can be a bit higher. Skill cores have the same cache as the M1 but with other architectural improvements (most likely, higher peak clock speeds and faster memory access, such as the A15).
Together, Apple says it gives the M2 18 percent more total CPU performance than the M1. This means you can probably expect Geekbench 5 CPU test results somewhere around 8,800-9,000. That’s about 20 percent or less of the power consumption of a Ryzen 7 3800X desktop CPU.
As a reference to recent mobile low-power chips, Intel’s best current ultraportable CPU, the Core i7-1255U, scores just under 7,000 when using more than twice as much power. Of course there are faster laptop processors from both Intel and AMD, but they consume dramatically more power and are better compared to the M2 Max or M2 Ultra.
M2 vs M1: GPU
The A15 has made an interesting change from the A14 before. The number of GPU cores has increased from 4 to 5, but that 5th core is only capable of iPhone 13 Pro and Pro Max. The M2 doubles the GPU core from A15 (as M1 did above A14), and so it is available with up to 10 GPU cores. The entry-level MacBook Air, however, has only 8 GPU cores enabled.
With only 25 percent more GPU cores, one can expect maximum 25 percent better GPU performance. But a lot of graphics performance is limited by working memory bandwidth, and the M2 makes a big improvement there. With other modifications (possibly a higher maximum clock speed enabled by the refined manufacturing process), Apple says GPU performance is up to 35 percent better than the M1, but power consumption has also increased slightly. In fact, Apple’s chart highlights 35 percent performance gains but shows almost the same amount of power gains. However, Apple claims that the M2 chip delivers 25 percent better performance at the same power level as the M1.
25 percent more core and 25 percent more performance. Who cares?
Still, it looks like a highly power-efficient GPU. Apple likes to compare the wild, saying that it is 2.3 times faster than the Intel Core i7-1255U at the same power level. We can’t help but notice that Apple hasn’t compared its chip to an AMD’s ultraportable laptop chips, which are generally superior to Intel in that category.
Either way, it doesn’t matter. Graphics performance improvements are always welcome, but 3D graphics performance is one area you’ll only compare to a Windows PC because it affects content creation apps like Adobe Premiere. Comparing processors from different architectures is always questionable, and it doubles down on trying to generalize “graphics performance” to include GPU computing, real-time 3D rendering, and more.
Suffice it to say that the M2 has a 25 percent larger GPU and is much faster, just like the A14’s A14. You see a pattern here, right?
M2 vs. M1: Memory
The memory subsystem is an area where the “M2 is A15 to M1 and A14” pattern is not true. The A14 and A15 have the same main memory subsystem space: a 64-bit wide LPDDR4x interface. The M1 doubles it to 128 bits wide and sticks with the LPDDR4x, for a maximum of 68.25GB / s, the M2 switches to LPDDR5x, which increases the memory bandwidth to 100GB / s.
Note that the M1 Pro switched to LPDDR5x, a 256-bit memory interface gives it 200GB / s bandwidth. The M1 Max has doubled to a 512-bit interface for 400GB / s.
So, while M2 has 50 percent more memory bandwidth than M1, it doesn’t Not necessarily Keep in mind that M2 Pro and M2 Max will have 50% more memory bandwidth than M1 Pro / Max. To do this Apple needs to build a wide memory bus, which is possible but expensive and complex.
In addition to the big boost in memory bandwidth, the M2 is available with up to 24GB of RAM. The M1 only offers 8GB and 16GB configurations, both still available on the M2.
Although this is not stated directly, we are willing to bet that the system-level cache (an on-chip cache that is accessible on each side of the chip) is 50 percent larger than the M2. Apple has increased the size of this cache from 16GB in A14 to 32MB in A15. The M1 features 24MB of the SLC, and it is reasonable to expect that it has doubled to 48MB in the M2. We can’t say until the M2 MacBooks ship in July and can be thoroughly tested, but the size of the SLC cache is a big deal. It helps EverythingSince it is accessible to CPUs, GPUs, neural engines, media encoders, you name it.
M2 vs. M1: Other improvements
Of course, Apple is quick to point out that the M2 is more than just a CPU, GPU and RAM. It includes a complete system-on-chip and hardware acceleration block important for the advanced features that Apple is adding to MacOS and iPadOS. The neural engine, although still 16 cores, can perform 15.8 trillion operations per second, more than 40 percent more than the 11 trillion operations of the neural engine in the M1. It’s just like jumping from A14 (11 trillion) to A15 (15.8 trillion).
The Media Engine, which is responsible for accelerating encoding and decoding audio and video, has also been upgraded to a new engine from the A14 to the A15. It supports H.264 and HEVC encoding and decoding up to 8K resolution and includes ProRes video acceleration support. Interestingly, the M1 does not have this version of the Media Engine, but the M1 Pro and M1 Max do. Originally, the M1 was shipped with a media engine from the A14, and the M1 Pro, Max, and Ultra, as well as the M2 have a media engine from the A15 (two on the M1 Max and four on the M1 Ultra).
It is possible that the Media Engine will be upgraded to A16 again (probably with AV1 hardware acceleration) and M2 Pro, Max and Ultra will get an upgrade to that version.
All in all it looks like a nice upgrade over the M2 M1, though not revolutionary. It did exactly the same for the M1 A14 by applying a thin-and-light laptop class chip to improve all the architecture we saw in the A15. And since the M1 processor (M1 Pro, Max and Ultra) has advanced like our LPDDR5x memory and new media engine, it’s not a clear win across the board.
It will be interesting to see Apple’s evolution in the inevitable M2 Pro, M2 Max, and M2 Ultra. We suspect that they may have been built using TSMC’s more advanced 3nm process technology, and when the iPhone 14 Pro was announced later this year we saw some improvements in the A16 – probably an advanced media engine and probably a Also the new neural engine. The M2 is a great start for Apple’s second-generation Apple Silicon for the Mac (and iPad Pro), but not surprisingly.