Understand two top chips in one article - Can Apple A13 beat Kirin 990?

For many technology fans, Apple's autumn conference is like a festival. After this conference, it is still expected to be flooded with screens. In addition to the mixed reception camera, another huge highlight of this conference is the Apple A13 processor.

At the press conference, Apple rarely compared its own machine with Huawei P30 Pro, but the key point was that Apple used its previous generation A12 chip to "crush" Huawei P30 Pro. Some netizens used the picture below to tease Huawei.

Of course, the reality is not like this, because Huawei's latest Kirin 990 chip has not yet been released, and Apple can only compare with "old phones". Such a comparison makes the loser very unhappy, and the winner also feels that the victory is unfair. Today, the editor will take you to take a look at the differences between Apple A13 and Kirin 990, two of the world's top chips, for a fair showdown.

What is Kirin 990?

First, let's take a look at Kirin 990. Kirin 990 has two versions, 4G and 5G. The main difference is whether it integrates 5G communication baseband. The 5G version supports two communication modes, independent networking and non-independent networking, and is at the forefront of the world. This is the biggest difference between the two versions. The following is some information about Kirin 990 (5G) for readers to ignore (please skip to the next paragraph):

In terms of technology, the 7nm+ EUV process is used, and the board area is 36% smaller than other solutions in the industry. This is the world's first mobile terminal chip with more than 10 billion transistors, reaching 10.3 billion transistors, an increase of 4.4 billion transistors compared to the previous Kirin 980. In terms of CPU, the Kirin 990 5G uses 2 large cores (developed based on Cortex-A76) + 2 medium cores (developed based on Cortex-A76) + 4 small cores (Cortex-A55). Compared with the mainstream flagship chips in the industry, the single-core performance is 10% higher and the multi-core performance is 9% higher. In terms of GPU, the Kirin 990 5G is equipped with a 16-core Mali-G76 GPU. Compared with the mainstream flagship chips in the industry, the graphics processing performance is 6% higher and the energy efficiency is 20% better. In terms of NPU, Kirin 990 5G uses Huawei's self-developed Da Vinci architecture NPU, which adopts NPU dual large core + NPU micro core computing architecture. In the application scenario of face recognition, the energy efficiency of NPU micro core can be increased by up to 24 times compared with the large core.

Of course, you can't believe all of these materials. Reading an article is like listening to a person talk. You should not only listen to what the other party says, but also what the other party doesn't say. You can't just look at the official posters when looking at a chip. The author believes that the pros and cons of a mobile phone chip should generally be judged from the following four aspects: 1. Communication. For mobile phones, making calls and being able to use mobile networks anytime and anywhere are major factors; 2. CPU and GPU performance. Performance is the primary productivity, and there is no need to be afraid of this; 3. Architecture. Chip architecture is the main factor that determines the balance between mobile phone energy consumption and performance; 4. NPU. With the development of artificial intelligence, AI is no longer just a cloud-based thing. Just like the emergence of GPUs replaced part of the CPU's image processing functions, NPU is about to become the next measurement indicator, so the author includes it here.

First, communication. In terms of 5G communication, Huawei can be said to be the leader, with both base stations and basebands. Kirin 990 is equipped with the world's first Balong 5G baseband, but there are also two problems. First, the global 5G commercial use is not yet mature. As a leading 5G country, China is only piloting in some cities. At present, the three major operators have only 26 pilot cities, and these 26 cities will be equally divided by the three major operators. Before buying a mobile phone, you must confirm which operator's pilot city you are in. For most second- and third-tier cities, 5G is unnecessary. Secondly, due to the incomplete construction of 5G base stations, there is a huge heating problem when using 5G mobile phones. According to the author's previous test, when using 5G network to download, the heat of the mobile phone will increase sharply, and the heat increase can be clearly felt by holding it. The battery life of 5G mobile phones will also be shortened to 60%-70% of the original. Therefore, from these two aspects, 5G is not a good choice at present.

Second, architecture: Many friends don't pay much attention to architecture. In fact, architecture determines the balance between performance and power consumption of a mobile phone. What a mobile phone needs to consider is portability. A mobile phone that needs to carry a power bank is a complete failure for consumers. This is also the main reason why Jobs gave up x86 and made a version of ARM architecture. Although the x86 architecture has strong performance, its high power consumption makes many mobile phone manufacturers unable to afford it. Kirin 990, like its previous generation, adopts the Cortex-A76 architecture, and A77 has been released long ago. In response, Huawei said that the A77 architecture was not used because the power consumption of A77 was too high. Huawei would rather let consumers have a mobile phone with a better battery life, rather than focusing on improving the running speed and sacrificing battery life. On the other hand, the author believes that because Kirin 990 adopts a 5G baseband, it will greatly increase power consumption. Coupled with the high power consumption of A77, it is simply a catastrophic heat for Huawei mobile phones, which also forces it to choose the A76 architecture with lower power consumption.

Third, performance: In terms of performance, Kirin 990 fine-tuned the previous Kirin 980's 4 large cores and 4 small cores design, using 2 large cores (developed based on Cortex-A76) + 2 medium cores (developed based on Cortex-A76) + 4 small cores (Cortex-A55), the large core frequency is 2.86GHz, higher than the previous generation 980's 2.6GHz, the medium core 2.38GHz is lower than the previous generation, and the small core frequency is 1.95GHz, higher than the previous generation 1.8GHz, so according to the author's guess, this generation of 990 has little improvement in performance. According to the data released at the Huawei conference, compared with Qualcomm Snapdragon 855, the single-core performance increased by 10%, and the multi-core performance increased by 9%. There will be detailed tests below.

Fourth, NPU: Kirin 970 is the world's first mobile processor equipped with NPU on the end. At that time, it was equipped with the Cambrian 1H intelligent processor. It was first launched on Huawei Mate20 and achieved amazing performance. Many black technologies were released in image and text processing. Two generations later, in Kirin 990, Huawei switched to the self-developed Da Vinci architecture. AI has always been Huawei's strength. The score of Kirin 990 on AI BenchMark reached 476% of Kirin 980, which shows its powerful AI capabilities.

AI performance enhancement is the trend of both

Regarding the AI ​​capabilities of mobile phones, many people think that AI is a matter of servers and has nothing to do with terminals. However, everyone still remembers the amazing performance of Huawei's Mate20 series. Huawei Mate20 Pro uses Kirin 970, which was equipped with Cambrian 1H chip at the time. It is also the world's first mobile phone chip equipped with NPU. This may be a bit unfair to Apple, because Apple has already had relevant AI capabilities, but it is called the Bible Network Engine. Let's talk about the amazing performance of Kirin series NPU. First of all, in terms of imaging, starting with 970, Huawei's flagship mobile phones can realize AI portrait color retention, real-time portrait recognition and other functions. In terms of this kind of black technology, Huawei users can be said to have had enough.

In addition, in terms of portrait recognition, such as night scene intelligent supplement, low-light shooting, skin smoothing, etc., NPU plays a role. In terms of text recognition, Huawei phones equipped with NPU can instantly translate text into foreign languages ​​and keep the font similar to the original text. As shown in the figure below, a Chinese paragraph is translated by Huawei P30 Pro.

Let's talk about Apple. Starting with A11, Apple has equipped its own A series chips with a hardware dedicated to machine learning - the "neural engine" actually refers to a processor package encapsulated together. This computing package is professionally called Soc (System-on-a-Chip). According to different divisions of labor, many processing units with special functions are added. For example, the GPU we are most familiar with is now the number of independent units in this package has been increasing, such as ISP (image processing), Modem (communication module), DSP (digital signal processing), etc. Different data coming in and being handed over to computing modules with different specialties for processing will get better results and higher energy efficiency.

To some extent, Apple's "neural network engine" and Huawei's NPU have the same function, but at the beginning, Huawei's Cambrian NPU attracted more attention, while Apple added its own things and rarely talked about them. The "neural network engine" and NPU have roughly similar functions, and their image capabilities are greatly enhanced. Huawei's NPU is more reflected in the camera, while Apple's is relatively more reflected in face recognition. So far, Apple's face recognition function is the best, and this is the reason. Among them, for the "Face ID" of iPhone X, the Face ID function will project more than 30,000 infrared light spots that are invisible to the naked eye, and then transmit the obtained infrared image and dot pattern to the neural network to create a mathematical model of the face, and then send this data to the secure partition to confirm whether the data matches. If the appearance changes over time, it can also adjust and adapt accordingly.

Judging from the amazing performance of the two giants, AI is increasingly becoming a major trend in future mobile phones and will gradually become an indispensable part of future smartphones.

In addition, it should be added that the single GPU core area of ​​A12 is reduced by 37% compared with A11. Obviously, with the support of the new 7nm process, A12 can add an extra GPU core while further reducing the area. The area of ​​the neural network unit has increased by 3.16 times, and the performance has skyrocketed. A13 is an upgrade based on A12, and its performance is of course more powerful.

The big competition between Apple and Huawei

Next, of course, comes the most important part of this article. Which one is better, the Apple A13 or the Kirin 990? Apple, which has always only talked about user experience, opened the benchmark for the first time at the press conference and compared it with its competitors. Does this mean that mobile phones have entered the benchmark era from now on, or has the always arrogant Apple finally "not enough innovation but enough benchmark"?

At this conference, although Apple compared the running scores with its competitors, it can be seen from the comparison chart that the performance of the iPhone 11 equipped with the A13 chip far exceeds that of the Huawei P30 Pro equipped with the Kirin 980. So which one is stronger when compared with the Kirin 990?

The iPhone 11's running score records were leaked on the famous CPU running score website Geekbench, with a single-core score of 5415 points and a multi-core score of 11294 points, which provided an important reference for the author's comparison plan.

At the Kirin 990 launch conference, the data released were all based on Qualcomm Snapdragon 855: compared with Snapdragon 855, Kirin 990 has 10% higher single-core performance, 9% higher multi-core performance, and 6% higher graphics processing performance. The author checked Geekbenchmark, a more authoritative benchmark website in the industry, and found that the single-core score of Samsung S10+ is 4461 points, and the multi-core score is 10003 points. In the Android camp, the performance of Samsung S10+ is obvious to all. According to the performance improvement ratio announced by Huawei, the Kirin 990 scores as follows: single-core 4907 points, multi-core 10901 points. According to this standard, Kirin 990 loses to Apple A13, but the gap is not very large.

Based on the above inferences, we can roughly conclude that:

1. Apple A13 chip performance is better than Kirin 990;

2. The gap between Apple A13 and Kirin 990 is not very huge. The fact is not that "Apple has performance, Huawei has baseband" as circulated on the Internet. In terms of performance, the gap between the two is not wide;

3. The A76 architecture adopted by Kirin 990 has limited impact on performance. After the process enters 5nm, the next generation of Kirin chips has a lot of room to catch up.

It should be noted that the above comparison is not very scientific, and there is a certain error in the calculation of the running points, so only a rough conclusion can be drawn. In addition, the same chip will have different performance on different machines. The specific performance of the two will have to wait until the release of Huawei Mate30 Pro before an actual comparison can be made.