This article is from WeChat official account: Old Stone talk about the core (ID: laoshi_tanxin) , Author: Old stone, the original title:” Apple M1 Chip: How to Start an Era”, head picture from: Visual China
On November 11th, the trilogy of Apple’s press conference this year finally came to an end. In this finale, the first generation of Apple’s self-developed chip, Apple Silicon, was released: the M1 chip. The first batch of Mac machines equipped with M1 chip has three models: MacBook Air, 13-inch MacBook Pro and MacMini. Whether it is the performance improvement announced at the press conference or the actual evaluation of these machines in the past few days, the performance and power consumption of the M1 chip have completely exceeded people’s expectations, and it is not an exaggeration to describe it as “subversion”.
Today’s article will talk about Apple’s M1 chip. From the perspective of chip design, I would like to take a look at the three main reasons why the M1 chip is so awesome.
PPA optimization of chip design
When designing a chip, one of the most important principles is to optimize the PPA, that is, to optimize the power consumption of the chip as much as possible.(Power), performance (Performance) and area (Area) . Under normal circumstances, these three points cannot have both.
For example, in order to improve the performance of the chip, we can add multi-stage pipelines, increase the bus width, or increase various hard-core processing units, but at this time, it is very likely to pay higher power consumption and larger Chip area as a price. On the contrary, if we want to design low-power chips, it is very likely that we need to sacrifice some chip performance.
So in actual engineering practice, power consumption, performance, and area are often trade-offs and balances. For a chip, there is no perfect design, only perfect balance. It’s like buying food for you with a limited budget, it’s difficult to buy bothFish, I bought bear paws again.
But this time the release of the Apple M1 chip gives people the most intuitive feeling that you can have both fish and bear’s paws under the same budget.
Apple released this picture at the press conference. It compares the M1 chip with the CPU of a certain notebook, but Apple did not say who it was. For the convenience of description, let’s call it “Zhang San Chip”. In this graph, the horizontal axis is power consumption and the vertical axis is performance. It can be seen that for a given power consumption standard, the performance of M1 is twice that of Zhang San. For a given performance requirement, the power consumption of M1 is a quarter of Zhang San.
In other words, in the two dimensions of high performance and low power consumption, which tend to trade each other, the M1 chip has made great improvements at the same time. In addition to the CPU, the GPU in the M1 has also achieved similar performance improvements and reduced power consumption. Apple said that M1 has the best CPU performance per watt in the world and the fastest integrated graphics card in the world.
In the previous articleIn “Intel 11th Generation Core TigerLake: The Most Sharpest Comment on the Internet”, it is said that no matter what marketing technique is used, running scores will never be bypassed. Although Apple did not officially mention the running scores, various evaluations have given quite high running scores.
Judging from Geekbench’s running scores, the three products equipped with the M1 chip this time, regardless of single-core or multi-core, ran out quite high scores:
Mac Mini: 1682 / 7097
MacBook Air: 1687 / 7433
MacBook Pro: 1714 / 6802
In fact, their single-core running scores surpass AMD’s just released Ryzen 9 5950X, and also exceed Intel’s TigerLake flagship version, which is the 11th-generation Core i7-1165G7.
In addition, the single-core and multi-core running scores of these three products also exceed those of the 2019 16-inch MacBook Pro currently on sale. In other words, from the CPU score alone, this new macbook air is faster than the flagship notebook Apple just released last year.
Not only the performance, but also the battery life of notebooks equipped with the M1 chip, which can reach up to 20 hours.
In my opinion, there are three main reasons for the M1 chip to achieve such a huge energy efficiency improvement: One is the use of 5nm process, the second is some chip architecture innovation, and the third is software and hardware. Deep optimization, let’s talk about them one by one.
TSMC 5nm process
Many articles did not mention the manufacturing process, or did not pay attention to it. In fact, advanced technology plays a very important role in improving the energy efficiency of chips. Because of the improvement of chip performance in the past ten years, more than 60% have directly or indirectly benefited from the improvement of semiconductor technology, while only 17% came from the upgrade of chip architecture. One of the main reasons why AMD can “overtake” Intel is to abandon GlobalFoundries and turn to TSMC.
M1 should be the world’s first and only notebook processor chip using TSMC’s 5nm process, which contains 16 billion transistors. According to TSMC’s data, compared with the previous generation of 7-nanometer process, transistors manufactured using the 5-nanometer process: density increased by 80%, speed increased by 15%, and power consumption reduced by 30%. With the new manufacturing process, more transistors can be packed into a chip while the chip area remains the same, and these transistors have lower power consumption and higher performance.
TSMC process roadmap (picture from WikiChip)
In other words, even if Apple does nothing, simply tape out the A13 chip with a 5-nanometer process, ideally, it can achieve these “free” performance, power consumption and area optimization. Of course, the actual situation is much more complicated than this.
Be aware that the 2019 16-inch MacBook Pro uses Intel’s 9th-generation CPU and uses Intel’s 14-nanometer process, which is at least two generations away from TSMC’s 5-nanometer process. So it is not difficult to understand why, from the point of view of running points, the MacBook Air equipped with M1 will reduce the dimension and hit the 16-inch MacBook Pro.
Apple M1 package drawing (picture from Apple)
chip architecture innovation
What I want to talk about here is Apple’s UMA structure, which is the unified memory architecture shown this time. A big principle for designing chips is that the closer the place where the data is stored is to the place where the data is used, the higher the performance and the lower the power consumption. So Apple integrated the memory particles originally on the circuit board into the chip package.
The main advantage of this is that the CPU, GPU, and AI engine on the chip can access the memory faster, while also greatly reducing the power consumption of data transmission. In addition, memory can be shared between modules, which saves a lot of data transfer and copy overhead.
Apple UMA structure diagram (picture from Apple)
It is worth noting that this architecture design and packaging method is not unique to Apple. In fact, similar methods are used in Nvidia’s A100GPU, AMD’s Rome processor, and Intel and Xilinx’s high-end FPGA chips. The method can integrate multiple different computing and storage units in the same chip package.
There are many specific implementation methods for this packaging method, such as chiplets used by AMD, Intel’s EMIB technology, and SSI used by Xilinx in FPGAs, which is stacked silicon interconnect technology. and many more. The industry refers to these technologies as 2.5D packaging technology, that is, connecting multiple small silicon chips in the horizontal direction to form a large chip. For specific analysis of EMIB and SSI technology, you can read the two articles I wrote before ” 2019 Annual Meeting of the bumper 10 nm technology do 》.
Foveros 3D package diagram (picture from Intel)
Collaborative optimization of Apple ecology
The third reason why Apple’s M1 chip has achieved a significant increase in energy efficiency is the deep optimization of Apple’s software and hardware. I think this is also Apple’s biggest competitive advantage. This is because other manufacturers may also (or necessarily) master the 5-nanometer process mentioned above, or the upgrade of chip architecture and packaging technology. And adopt these technologies. But only combining software and hardware, operating system and ecology to do deep optimization is unique to Apple. And this is the reason why some and only Apple can sell ARM-based CPUs as awesome products.
Some people say that the launch of Apple’s M1 chip proves that the Wintel alliance is not difficult to break. I don’t agree with this point. Consumer-grade CPUs based on the arm architecture have been said for many, many years, but looking at the current software and hardware manufacturers other than Apple, who can meet this task? Intel amd will definitely not do this kind of self-abolition of martial arts. Needless to say, Microsoft has tried to do arm-based ecology before, but it ended in failure. So let Lenovo and Dell do arm notebooks, or let Qualcomm and NvidiaIt is very unrealistic for non-cpu vendors to do ecology.
As for the collaborative optimization of software and hardware, I very much agree with the answer of some big v wood dragon. You can check it out. As long as the PC software and hardware vendors are fighting separately, as long as different software and hardware need to be compatible and adapted to each other, whether it is an operating system, development tools, drivers, CPU, GPU, memory, etc., this ecosystem is difficult to crack. Sometimes the ecosystem is like a snowball rolling down from the top of a mountain. Once you are caught in it, it is difficult to escape by your own strength, let alone change its trajectory by your own strength.
In contrast, Apple has its own ecology. The release of the M1 chip this time has officially completed the closed loop of this ecology. You can say that the Apple ecosystem is too closed, or there are problems of this kind, but only this closed ecosystem can produce things like M1 chips.
From this picture of Anandtech, it can be seen that in the past 5 years, the single-thread performance of Intel chips has increased by 28%, while Apple has increased by nearly three times. Therefore, it is an inevitable choice for Apple to adopt the self-developed Apple Silicon chip.
(Picture from AnadTech)
However, unlike Intel AMD, you cannot buy this M1 chip alone. In order to use the high energy efficiency of M1, it is necessary to buy the entire ecology of Apple. And even for a closed-loop ecosystem like Apple, it will take two years to fully switch to the arm architecture. There will be a lot of ecological migration work involved, and it will inevitably cause a lot of additional work pressure on developers. So this ecological transformation is not easy.
Conclusion
The release of Apple’s M1 chip also represents the re-entry of consumer processors based on the arm architecture on the stage of history and the beginning of a head-on approach to traditional processors based on x86 architecture. Although it is still too early to say that the x86 architecture is replaced by arm, in many fields, such as content creation fields such as video and image that mac computers are good at, the M1 chip has officially subverted people’s perception of high energy efficiency. I believe that as more follow-up products appear, there will be more disruptions. At the same time, how the x86 camp is going to fight is also the focus of attention.
This article is from WeChat official account: old stone talk about the core (ID: laoshi_tanxin) , author: old stone