The public version of ARM architecture is the king, and independent CPUs are useless?
|
For mobile phone processors, you don't need to pay too much attention to the number of cores and the main frequency, but should focus more on the architecture and process. It is very easy to judge the quality of the process. Basically, the smaller the number of nanometers, the more advanced it is. FinFET transistors are better than traditional 2D transistors. But it is more difficult to judge the quality of the architecture, because there are public versions and independent architectures. If you understand the naming rules of the public version architecture, you can probably know one or two. Because the independent architecture is different from the manufacturer, it is very difficult to judge from the naming. These obstacles ultimately hinder our understanding of the pros and cons of a SoC. Is the independent architecture strong or the public version architecture? Which is stronger or weaker between different independent architectures? For these questions, let's start with the instruction set. The game between reduced instruction set and complex instruction set The hard program inside the CPU used to guide calculations and optimization is called the "instruction set". It is the lowest-level instruction that the CPU can directly recognize, which is divided into complex instruction set and reduced instruction set. The complex instruction set is to improve the computer's execution speed by setting some complex instructions and changing some commonly used functions that were originally implemented by software to hardware instruction systems. Intel's famous X86 architecture is a typical complex instruction set product. In the early days of computers, when components were expensive, the main frequency was low, and the running speed was slow, this could greatly improve processing efficiency. However, as the complex instruction set became increasingly complex, this structure became larger and larger, and the versatility and running speed began to deteriorate, so another idea-driven reduced instruction set was born. The idea of the reduced instruction set is to simplify the computer instruction function, reduce the average execution cycle of the instruction, and implement more complex functions with a subroutine, thereby increasing the computer's working main frequency. At the same time, a large number of general registers are used to increase the execution speed of the subroutine. The ARM architecture of ARM and the MIPS architecture of Imagination Technologies both belong to this system. Almost all the popular mobile processors currently use the ARM architecture. This reduced instruction set architecture brings four major advantages: first, small size, low power consumption, low cost, and strong performance; second, a large number of registers are used and most data operations are completed in registers, so the instruction execution speed is faster; third, the addressing method is flexible and simple, and the execution efficiency is high; fourth, the instruction length is fixed, and the processing efficiency can be improved through multi-pipeline mode. The ARM architecture is also divided into multiple generations such as ARMv6, ARMv7, and ARMv8. The core designed based on the ARMv6 instruction set is ARM11, which is widely used in early smartphones, especially in Nokia's Symbian system phones. ARMv7 is the most used architecture in the new smartphone era. The Cortex-A7/A8/A9/A15 cores we are familiar with are all products of this architecture. The ARMv8 instruction set was released in November 2011. It supported the 64-bit instruction set for the first time in ARM history, forming the core foundation for Apple's first 64-bit processor A9 in 2013. Nowadays, the common mobile phone autonomous/non-autonomous processor architectures are all based on the arm instruction set (except for a few Intel core mobile phones that use the X86 instruction set). Why are mobile phone processors divided into public version/autonomous version? Although it has established a unified position in the mobile field with the ARMvX architecture, this company from the British Isles did not follow Intel's example of doing everything by itself, but opened up authorization to the outside world, allowing manufacturers to design and produce final products according to their own needs. As an authorization, ARM divides it into two types. The mainstream is to authorize ARM-designed IP cores, such as Cortex-A53/A72. After obtaining such authorization, manufacturers only need to select the number of cores, bus interconnection, and cache according to their needs to basically complete the design of the CPU part. So we call this kind of direct use of ARM-designed core solution a public version architecture, and chips from MediaTek, Samsung, and HiSilicon belong to this category. Another solution is to authorize ARM architecture, such as ARMv7/ARMv8. After obtaining these instruction set architectures, manufacturers have to design the kernel by themselves, and then complete the construction of the entire CPU part. This is what we call autonomous architecture, and most of Apple and Qualcomm's chips are the products of this solution. The biggest advantage of the first public version solution is that it saves time, effort and cost. Chip manufacturers only need to follow the pace of ARM, and they will not be left behind. They can not only guarantee performance, but also complete the listing of finished products in the fastest time. At the same time, it is also convenient for people who know a little about mobile phones to judge the strength and positioning of the chip. The disadvantages are that they lose differentiation and cannot form an exclusive selling point; the second is that they must always keep up with the pace of ARM. Once they fall behind, they will be easily perceived by the outside world and instantly become low-end and eliminated. The advantage of the second independent solution is flexibility. Manufacturers can design single-core performance with better than public version IP and lower power consumption. At the same time, they also have a high degree of freedom in bus interconnection and great room for development. However, this solution must be based on the premise that chip manufacturers can afford to spend money and hire people. It takes a long time to achieve results that surpass the public version. In essence, which solution to adopt is just a different decision made by chip manufacturers based on their own strength, financial resources, time cost and final product needs. Some manufacturers may always use the public version architecture, but after the relevant reserves are mature, they may turn to independent research and development, such as Samsung. Some manufacturers, although they have always loved their own architecture, will return to the public version architecture, such as Qualcomm, when the rhythm is disrupted or when they pay more attention to the cost of mid- and low-end chips. Nothing is immutable, so we don't need to spend energy on distinguishing the public version and the independent version of the architecture, as long as we focus on the actual performance. Actual performance cannot be finalized . Can the actual performance be represented by just looking at the architecture? In fact, it is not the case. 1. Public version . In terms of actual performance, it is very convenient to compare the public version architectures, which can be roughly summarized as A72>A57>A15>A17>A9>A53>A35>A8>A7>A5. Except for the 32-bit A15 and A17 cores (A17 is optimized on the basis of A12, with performance close to A15, but lower power consumption), the remaining cores are named with one digit for the 32-bit ARMv7 instruction set and two digits for the 64-bit ARMv8 instruction set. 2. Apple. But after adding the independent architecture, the comparison becomes complicated. Let's talk about Apple's independent architecture first. Apple has been at the forefront of developing its own core since the A6 processor. It first launched the Swift architecture based on the ARMv7 design. Its performance is between the public version of the A9 and A15, and is stronger than Qualcomm's Krait 300 of the same period. With the A7, Apple has demonstrated unprecedented design capabilities. In just one year, it designed the Cyclone core based on the 64-bit ARMv8 architecture, which is one year ahead of its competitors. The entire A7 chip integrates more than 1 billion transistors, and the performance of the dual-core configuration alone is equivalent to that of the quad-core A15 processor. When it comes to the A8 chip equipped with the iPhone 6, the improved Typhoon architecture improves the processor's performance by 25%. The single-core performance exceeds the A57, and the multi-core performance is only slightly behind the Exynos 7420 and Snapdragon 810 with eight cores A57+A53. As for the latest A9 chip, it uses the third-generation 64-bit architecture Twister core, and the CPU performance is 70% higher than that of A8. The single-core performance is ahead of the Kirin 950 using the latest A72 architecture, and it is still the strongest single-core core in commercial chips. 3. Qualcomm Another company that is keen on developing its own architecture is Qualcomm. As early as the Snapdragon S1 era, Qualcomm used the Scorpion core based on the ARMv7 architecture in QSD8250. Compared with the popular A8/A9 public version cores of the same period, Scorpion has added some out-of-order execution capabilities, supports asynchronous symmetric multi-processing, and has outstanding advantages in high main frequency, low power consumption and enhanced floating-point operations. The specific performance is slightly weaker than A9. This architecture has been used for three generations of Snapdragon S1, S2, and S3 processors, and it has become a bit old and weak in the later period, so Qualcomm launched the Krait core. The Krait core is divided into four generations, Krait 200, Krait 300, Krait 400, and Krait 450, in chronological order. They are all based on the ARMv7 architecture. The first generation of Krait is used in the Snapdragon S4 processor. It can perform three fetch and decode operations in one clock cycle. The back-end execution units have increased from three in the Scorpion to seven, and the pipeline has increased from 10 to 11. The actual performance is slightly weaker than the A15 with a 15-stage pipeline. The second-generation Krait 300 core has improved the branch prediction module, added an out-of-order execution engine, and brought better floating-point computing capabilities. It is used in the first-generation Snapdragon 600, with performance close to that of the A15, but lower power consumption. The third-generation Krait 400 is manufactured using the 28nm HPM process, with an improved memory controller, lower latency, and a higher-frequency secondary cache. Its performance is stronger than the A15. The Snapdragon 800/801 we are familiar with uses this core. The last generation of Krait 450 is used in the uncommon Snapdragon 805, and the main change is that the main frequency is raised to an exaggerated 2.5GHz. After Krait 450, Qualcomm's new generation of Snapdragon 810 turned to the public version A57 core for competitive reasons, and the performance was relatively stable. In the upcoming Snapdragon 820, Qualcomm finally launched the long-developed Kryo core, which is Qualcomm's first self-developed 64-bit core, and its single-threaded performance is stronger than the latest public version core A72. 4. Others In addition, Samsung and Nvidia have also been involved in the ranks of self-developed cores. Samsung's Orion chips have always used public version Cortex cores, and last year they also made a lot of headlines with the 14nm process Exynos 7420. However, for Samsung, which has always focused on research and development, it is certainly unwilling to just watch others play with independent architectures. Therefore, the Exynos 8890 equipped on the upcoming Samsung S7 uses the self-developed Mongoose core based on the ARMv8 architecture to replace the A57. The performance is also stronger than the A72 and comparable to Qualcomm's Kryo. It is worth mentioning that in the animal world, the mongoose referred to by Mongoose is the natural enemy of Qualcomm's previous Krait (Bunny Ring Snake). In addition to Qualcomm, NVIDIA also used its self-developed 64-bit Denver core on a version of the previous Tegra K1 chip, but due to late listing, power consumption and lack of baseband, there are almost no models using it on the market. The Tegra X1 launched later returned to the public version A57+A53 big and small core design. Summary After the above description and review, we can basically summarize the following rule: independent architectures are usually more powerful than the public version architectures of the same period, but it should be noted that this strong performance refers to single-core performance, while in terms of multi-core performance, the CPU with an eight-core design is undoubtedly much stronger. Although independent architecture has many advantages, it is not for everyone. Even a strong company like Qualcomm has reported that the next generation Snapdragon 830 will abandon independent architecture. Although I don't believe it is true, it does reflect the huge investment of resources required for independent design. As consumers, as long as the chips they sell are of the same level of performance, it is enough. There is really no need to be fussy about a few hundred extra points, and there is no need to criticize each other for whether they are independently developed. As a winner of Toutiao's Qingyun Plan and Baijiahao's Bai+ Plan, the 2019 Baidu Digital Author of the Year, the Baijiahao's Most Popular Author in the Technology Field, the 2019 Sogou Technology and Culture Author, and the 2021 Baijiahao Quarterly Influential Creator, he has won many awards, including the 2013 Sohu Best Industry Media Person, the 2015 China New Media Entrepreneurship Competition Beijing Third Place, the 2015 Guangmang Experience Award, the 2015 China New Media Entrepreneurship Competition Finals Third Place, and the 2018 Baidu Dynamic Annual Powerful Celebrity. |
<<: Serious security risks! 1 in 20 Android phones is rooted
Recommend
Taklimakan is not a hopeless situation? The birth of desert ninjas!
Few people can live here for long. Even in today&...
Invoca: 85% of customer interactions are handled without a human
Brands are betting on AI and digital communicatio...
AE animation class: The secret of keyframes - how to adjust the animation rhythm that makes people happy
: : : : : : : : : : : : : : : : : : : : : : : : : ...
Put your mirrorless camera aside and leave the professional photography to vivo X50 Pro!
Although vivo has added multiple product lines in...
What is Google overseas promotion and what are the advantages of Google promotion?
Google promotion, in layman's terms, is to us...
Analysis of Himalaya audio content operations!
The author has conducted an in-depth analysis of ...
How to arouse users’ desire for products?
When we hear about or see a very good new product...
What is the role and significance of shooting a company promotional video?
Company promotional video shooting is a common pr...
Is there any scientific basis for strictly controlling electronic products to protect eyesight?
In order to protect their children's eyesight...
Android is in chaos, Apple is king
Do you have the same or similar feelings: The And...
The secret code of Weibo "stalk" marketing traffic
Farewell and longing for the distant past, breaki...
Fish may contain veterinary drugs? How to buy fish safely?
Produced by: Science Popularization China Author:...
International Left-Handers Day | Are left-handed people really smarter?
August 13 is International Left-Handers Day, and ...
How can the education and training industry effectively attract new customers? Share 3 tips!
Difficulty in recruiting students and high costs ...
How to use Tik Tok app? What is the delivery method?
When opening a store on Douyin, merchants also ne...