The new Apple M1 and M2 chips have been surprising with their performance and energy efficiency. But what is behind them? Here you will understand everything you need to know about these two SoCs that have revolutionized the world of Mac and mobile devices from the Cupertino company.
Apple has gone through various ISAs or CPU architectures since its inception, such as the 6502 of the MOS Technology chips that it used in its beginnings, later going to Motorola’s 68000 or 68k, until landing on the successful PowerPC with which they have remained longer and created by the AIM alliance (Apple IBM Motorola).
After that, they decided to jump to x86-64 with the Intel chips they used until 2020, and then went on to design their Arm-based chips. But why Arm?
All this has meant that both Apple and the software developers for the platform have to create ports of their macOS and programs respectively to adapt them to the new architecture. Or, failing that, use emulators or dynamic translators like Rosetta. But the changes have always been worth it and have seen Apple gain noticeable improvements.
In the case of Arm, on the one hand, they already had a lot of experience, since they had already developed their A-Series SoCs for iPhone and iPad mobile devices. In addition, this RISC architecture allows the creation of small cores, which allow better use of the chip surface area, as well as being more efficient in terms of performance/consumption ratio.
It is currently being rumored that Apple might already be thinking about RISC-V, another ISA that is very similar to Arm in terms of performance/consumption and other characteristics, being also RISC, but it is open source, so it is not they would have to pay a license as they currently do, and they could obtain greater benefits in the sales of their processing units.
On the other hand, due to the instability of Arm, which has passed from one hand to another, and with a failed purchase attempt by NVIDIA, this ISA would not be the best, since any course change could cause the company to change. way to license this technology to third parties or cause other harm. For this reason, RISC-V seems like a solid bet for the future that many are already considering, including competition such as Qualcomm, Samsung, Intel, etc.
What is M-Series?
The M-Series is the chips that Apple is designing for its Macbook and Mac computers, as well as equipping other mobile devices with them, such as the iPad Pro, for greater performance. This series is an SoC, not just a CPU, but other elements that must be highlighted are integrated into the same chip.
The CPU built into the M-Series chip is a derivative of the ones Apple uses for its A-Series. That is, it is based on the 64-bit ARMv8 ISA, and uses a mix of high-performance cores called Firestorm and energy-efficient cores called Icestorm.
This gives you good performance without sacrificing too much battery when you don’t need full power, which is when efficient cores come into play. This hybrid core technique is not new, it was already used by ARM and called big.LITTLE, and now it has been expanded by other designs, including those of Intel.
Although the current GPU is designed by Apple, the truth is that initially, the Cupertino company’s SoCs used a GPU from Imagination Technologies, which was dedicated to designing GPUs for ARM. This GPU is called PowerVR and from it derives the current Apple ones. That is to say, although there is secrecy and hermeticism about these GPUs, it seems that they could be derived from those of IMG.
These GPUs have good performance, although they are not comparable to AMD’s or NVIDIA’s designs, which are much more powerful. However, with the improvements introduced by Apple and the use of several graphics processing cores, the performance is more than decent. However, do not expect miracles in gaming, although they do obtain good results in design programs such as Adobe Creative, among others.
Apple has done a lot of work on its GPU to improve its photography capabilities, although this is not the case in gaming and rendering, where it is still beaten by the competition.
On the other hand, Apple has followed a heterogeneous SoC design with other additional units called the Neural Engine. It is a kind of NPU intended to improve performance with applications that use AI or artificial intelligence.
These drives are also built on a multi-core architecture, with a neural engine that can perform trillions of operations per second. In addition, it has a machine learning engine that accelerates the matrix multiplication (Apple calls it AMX blocks) necessary in these apps.
The system-level cache, or SLC, is another of the modules built into Apple’s SoC. It is the cache memory of the different levels that are handled by the processing units. This has different designs and capacities in each generation of chips.
It refers to a powerful internal bus that interconnects all the units in the SoC quickly and efficiently, with high bandwidth, low latency, and being ability to have a large data transfer to avoid bottlenecks.
Of course, as is usual with modern drives, the MMU, or memory management unit, is also built into the chips. It is a unit prepared for DDR4 (in its first generation) and DDR5 (in the new designs), specifically for the low-consumption version, LPDDR4, and LPDDR5 respectively.
This part is very important since it is the one that interconnects the SoC with the RAM memory through its channels. And it must do it as quickly and with the lowest possible latency so that the data and instruction-hungry chip doesn’t miss information to process.
The RAM memory has not been integrated inside the chip but remains outside in DRAM chips that are integrated on the same PCB. In this way, the amount of memory that is integrated into one design or another can be controlled, have several versions on the market.
Of course, the SoC also integrates other units, such as video encoding and decoding engines, to help the CPU in this task so that it does not get overloaded and performance suffers.
Of course, there’s also DSP for advanced image processing, a controller for USB 4 or Thunderbolt ports, and a secure enclave to enhance security and encryption without overburdening the CPU either.
It is likely that in the new generations more units will continue to be integrated to support new technologies or interfaces.
The first M series chip is the M1 which was released in 2020. This new Apple design marked a before and after, not only because of the change from x86 to Arm but also because of the results obtained. Furthermore, it proved that it could perform well enough for the firm’s laptops and desktops, as some were skeptical about it. Therefore, they knew how to overcome their series A.
This M1 model has been designed with the following characteristics :
- 16 billion transistors made on TSMC’s 5nm technology. They were the first to use this node exclusively.
- 16 Neural Engine cores to accelerate AI and deep learning applications.
- 8 CPU cores of which 4x are high-performance Firestorm, and 4x are efficient Icestorm.
- 8 GPU cores. All of them are the same and supposedly based on IMG’s PowerVR.
- Support of up to 16 GB of unified, shared memory for the GPU, NPU, and CPU.
- Memory Bandwidth 100 GB/s.
The M1 could be found in various configurations based on active cores, depending on which ones failed during manufacturing or not. And this can be used for more powerful versions of PCs or for devices like iPad tablets, which don’t need as much performance and may have some inactive cores.
In addition, other SoCs such as Pro, Max, and Ultra appeared that have more or less processing cores.
The new M1 Pro was an improved model to get some extra features. It is an M1 with a higher core count of processing units. Among the most notable features are:
- 33.7 billion transistors manufactured on TSMC 5nm node.
- 16 Neural Engine cores to accelerate AI applications.
- 10 CPU cores with 8 high-performance Firestorm cores and 2 smaller, lower-performance, less-powerful Icestorm-like cores.
- 16 identical GPU cores for higher performance over the base M1, achieving double the performance.
- Support of up to 32 GB of unified memory, shared by all processing units.
- Memory Bandwidth 200 GB/s.
This chip also has versions with both 10 cores and 8 active cores, so it can be found in those two variants.
The Max is another more powerful version of the M1 and that surpasses the M1 Pro, for those who demand more performance. The characteristics of this chip are as follows:
- 57 billion transistors made with TSMC’s 5nm node.
- 16 Neural Engine cores to accelerate AI applications.
- 10 CPU cores, in this sense it is the same as the Pro, with 8 Firestorm-type cores for higher performance and another 2 Icestorm for greater efficiency.
- 32 GPU cores, that is, twice as many as the Pro, further improving graphics processing.
- Support up to 64 GB of unified memory.
- Memory Bandwidth 400 GB/s.
|Characteristic||M1||M2||M1 Pro||M1 Max||M1 Ultra|
|Total CPU processor cores||8||8||8 / 10||10||twenty|
|High-efficiency CPU cores||4||4||two||two||4|
|High-performance CPU cores||4||4||6 / 8||8||16|
|GPU graphics cores||8||10||14 / 16||24 / 32||48 / 64|
|Maximum RAM memory||16 GB||24GB||32GB||64GB||128GB|
|RAM memory speed||50GB/s||100GB/s||200GB/s||400GB/s||800GB/s|
Finally, the top of the range arrives, the M1 Ultra, it is the most powerful of the entire M1 series. And its features are truly amazing, highlighting:
- 114 billion transistors made with TSMC 5nm node.
- 32 Neural Engine cores, double that of previous designs, enabling better performance in AI applications that demand integer and matrix operations.
- 20 CPU cores, with 16 high-performance Firestorm cores and another 4 high-efficiency Icestorm cores.
- 64 GPU cores, doubling the graphics capabilities of the Max.
- Support up to 128 GB of unified memory.
- Memory Bandwidth 800 GB/s.
In this case, you can find GPUs with 48 or 64 graphics cores, depending on the configuration.
|CPU||4 high-performance cores and 4 high-efficiency cores||4 high-performance cores and 4 high-efficiency cores|
|GPUs||8 cores (2.6 TFLOPS)||10 cores (3.6 TFLOPS)|
|Neural Engine||16 cores (11 trillion operations per second)||16 cores (15.8 trillion operations per second)|
|RAM||up to 16GB||up to 24GB|
|memory controller||LPDDR4-4266 | 8x 16CH | 68GB/sec||LPDDR5-6400 | 8x 16-bit CH | 100GB/sec|
|Encoding/Decoding||4K | H.264, H.265||8K | H.264, H.265, ProRes, ProRes RAW|
|transistors||16 billion||20 billion|
After the M1 Apple worked to bring the M2 in record time. This other high-performance chip has been the last to arrive and promises to improve those seen with the first generation. Regarding the technical characteristics of this chip, they stand out:
- 20 billion transistors manufactured on TSMC’s 2nd Gen 5nm node where manufacturing process optimizations have been made to extract greater performance from the chips.
- 16 next-generation Neural Engine cores with improvements from 11 trillion operations per second to 15.8 trillion operations per second.
- 8 CPU cores, 4x Firestorm, and 4x Icestorm for high performance and high efficiency respectively. However, these are not the same as the M1 and its variants, but are of a new generation, with an improved microarchitecture.
- 10 GPU cores, again from a new improved architecture over the cores in the M1. They are capable of developing 3.6 TFLOPS of performance compared to the 2.6 TFLOPS of the previous generation, which means 1,000,000,000,000 more floating point operations every second.
- Support of up to 24 GB of unified memory with improved drivers for higher bandwidth and lower latency.
- Memory Bandwidth 100 GB/s.
As was the case with the M1, the M2 also comes in various configurations depending on the active cores, being able to find chips with 8+8 CPU/GPU cores or 8+10.
M1 vs M2: differences and improvements
The new M2 has been a step forward concerning the M1. However, due to the increased number of cores in the Pro, Max, and Ultra editions, it has fallen behind these others until the Pro, Max, and Ultra versions of the M2 appear, if and when they appear before. of the M3.
It is surprising in the performance tests to see that in the multi-core Geekbench benchmark, notable differences are achieved between all the previous processors, the fastest being the M1 Ultra, given its greater number of cores, as expected. On the other hand, in the single-core tests, the differences are smaller between the Pro, Max, and Ultra, given their similarities. And in this case, it is surprising that the single-core M2 is a bit behind those editions of the M1, although considerably ahead of the M1.
According to tests, single-core performance has been improved by 11.5% over M1, while multi-core performance has been improved by up to 20% over M1. Figures that show that the M2 can leave really surprising figures when you see the derived versions Pro, Max, and Ultra, which should surpass all versions of the M1 in all tests.
Variants of the M2
For now, little is known about the Apple M3 chip or if there will be variants of the M2 before the M3 comes along. But variants of the M2 manufactured on TSMC’s first generation 3nm node and exclusively as has happened with the 5nm line (leaving the 3nm+ process for the M3) are likely to appear next year. For example, the units:
- M2 Pro: When it appears it will likely have 10 CPU cores and 20 GPU cores. Of them, some will be of high energy efficiency and others of low. Of course, it’s also likely to follow the same path as the M1 Pro and not add any more Neural Engine units over the base M2.
- M2 Max – In this case, the number of GPU cores will also be increased, probably leaving the same CPU cores concerning the M2 Pro, although varying the number of high-performance cores, which will grow compared to low-performance ones.
- M2 Ultra – again an increase in the number of CPU and GPU cores. With higher bandwidth and more Neural Engine cores than the previous ones.
- M3: It is possible that it will be the first to release the improved 3nm manufacturing node, and that it will integrate several CPUs and GPUs similar to the M1 and M2, as well as Neural Engine units, all of the new generation. In addition, it is most likely that this time, new functions will appear in front of the M2 as has happened with the M2 compared to the M1. For example, drivers for the upcoming Thunderbolt/USB4 replacement may appear.
The new M2 chips and their variants will go to the new Macbook Air and Macbook Pro, which likely won’t get any other upgrades beyond these new chips. In addition, the base M2 will also be used in the firm’s new high-performance tablets, such as the Apple iPad Air and the iPad Pro.
Read more about Top 10 Most Powerful Supercomputers In The World
Abram left his e-business studies to devote himself to his entrepreneurial projects. In 2017, he created the company Inbound Media and wrote articles about high-tech products for his Chromebookeur site. In 2019, Chromebookeur was renamed Macbound and became a general purchasing advice site. Today, Abram manages the development and growth of Macbound, surrounded by a young and talented team.