On the afternoon of August 29th, Huawei suddenly launched the Mate60 Pro. There was no advance announcement and the press conference was launched, and it was launched offline. This caught us off guard. The most important thing is the Mate60 Pro this time. It is also equipped with a very interesting chip Kirin 9000S. Don’t look at the Kirin 9000 in its name. In fact, this S has nothing to do with the original 9000. It is not an old stock chip. It is a brand-new build. What is more exciting is this The CPU and GPU parts of Kirin 9000S are the first to use Huawei's self-developed IP instead of completely using ARM's public version IP as before. The manufacturing process even uses an unpublished domestic process, which can be said to be very milestone. I have a product that makes me look forward to how it will perform? So I bought one from Shenzhen at the first time and came back to start the test as soon as possible. Let's take a look at how sacred this mysterious Kirin 9000S is! Kirin 9000S, as I said just now, is a chip that was released very suddenly but with a lot of headlines. Thanks to the help of an old acquaintance in Geek Bay, we have a glimpse of the internal structure of this core , but it is true. I can’t release the die shot for you to see because it involves some sensitive supplier information. What we can do is to draw a schematic diagram for you to explain the internal structure of Kirin 9000S. First, the CPU part of Kirin 9000S gave us a big surprise. Its small core uses the ordinary A510 , but the large core and medium core are not ARM public version IP or even a modified version , but a self-developed architecture like Apple. In fact, Huawei has been developing its own ARM processor core for a long time. But it has been only used on the server side before, that is, the Kunpeng processor, whose architecture is called Taishan architecture. This time, the Kirin 9000S uses a new generation of Taishan core that is the same as Kunpeng. Therefore, it inherits some server features such as super The thread is right. The Kirin 9000S is the first mobile ARM processor that supports hyper-threading. The specifications of the CPU are 8 cores and 12 threads. The CPU frequency is not high. The big core is 2.62GHz, and the middle core is 2.15GHz. It is a lot lower that the large core and the medium core use the same architecture , but from the figure, the large core area still takes up a bit more because the large core provides more L2 cache and may use a higher frequency library to build At the same time, the number of execution units of the large core may be a bit more. The entire CPU part shares 4MB L3 cache , and the system cache of the entire SoC is also 4MB. It is said that the cache is quite small, so let’s talk about this self-developed Taishan core in detail. It is a six-issue processor IPC. We use SPEC2017 to measure the level between A710 and X1 , but from the perspective of ROB, its scale is actually larger than X1 , but it is only due to reasons such as small cache and hyper-threading. The single-threaded IPC is still slightly lower than that of the X1. Fortunately, the four large cores of the Kirin 9000S all use this large core. It does not have a medium-sized core similar to the ARM A series. Then the problem comes. With the blessing of domestic manufacturing technology, what is the energy efficiency of this Taishan core? We tested it and felt that the situation is more complicated. If you just look at the highest frequency point, which is 2.6GHz, its performance does not seem to be optimistic. IPC has never played 888. At the same time, the power consumption of this X1 is more than 1W higher, which is obviously. It's not very good , but let's not jump to conclusions too easily. Will the performance be poor at 2.6GHz? Let's just test the mid-core and we will know that the energy efficiency of this Taishan mid-core is slightly inferior to the 8Gen1 A710 when it runs at 1.9GHz. Although this is still not very good , it is better than the big-core 2.6GHz. A lot of it gives me the feeling that the domestic process used this time may not be able to handle the high frequency 2.6GHz, which is considered high for it, but the energy efficiency of this process at low and medium frequencies should not be as bad as we thought. In order to further verify this domestic What is the level of the manufacturing process? I decided to use the A510 in this CPU to do an experiment. After all, the small core of the Kirin 9000S is the same as the 8Gen1 and 8+, which is A510. The energy efficiency of our smaller cores can roughly tell the process. Have you acted? Here I ran a stress-ng FFT at around 1.43GHz, that is, a fast Fourier transform test. Judging from the results, this domestic process is definitely far from the TSMC 4nm that produced 8+ , but at least in this In terms of low frequency, its energy efficiency does not seem to be worse than that of Samsung 4nm, and even the A510 of Mate60Pro can run at a slightly higher frequency than 8Gen1 under the same motherboard power consumption. Of course, the power consumption of small cores is too low, so there must be something in it The impact of various peripherals on the mainboard. Huawei’s optimization of the power consumption of the mainboard is considered relatively good, but at least it can be seen that this domestic process does not hold back too much. In terms of low-frequency energy efficiency, it can compete with Samsung’s 4nm wrist, which is already beyond me. It's expected , but it's still a problem that the high frequency cannot go up. I think this relatively wide Taishan architecture is a fire-fighting to a certain extent. It still has considerable performance under 2.6GHz . If Huawei comes up with a small-scale core If the high frequency does not go up, it will be a big trouble. After talking about the GPU next to the CPU part, there is no doubt that the self-developed GPU of Ma Liang 910 is a brand new architecture that we have never seen. Its specifications are quite large, with a total of 4 sets of CU. 1024 ALUs should be said to be stacked with some materials. The frequency is not too high, the highest is only 750MHz. We are also curious about how Huawei's first-generation self-developed GPU can perform. In addition, this SoC also integrates 4.9999 G's Baron baseband also self-developed a new generation of Da Vinci NPU and a fairly large ISP, etc. This specification really makes people look forward to it. It should be said that everything about the Kirin 9000S is too mysterious . I can't even guess what level it is. Now that we have passed the previous IPC and energy efficiency tests, we finally have a little understanding of this self-developed architecture and new process technology, but these things are too microscopic. I know that many viewers just come to see it. Everyone who wants to know where the Kirin 9000S is in all mobile phone chips ? What level is it? Okay, let’s satisfy this curiosity and run the Geekbench 5 CPU test first. We tried it in the extreme situation. Under the hyper-threading blessing, we can run to a single-core 1005 points and a multi-core 4019 points. In terms of this limit performance, it is still quite high . Today's flagship platform is far behind , but still surpasses Dimensity 8100, Snapdragon 888, and 8Gen1. The performance is also better than Huawei's own Kirin 9000. This level is definitely enough , but everyone knows that no matter how high the limit performance is, in fact . The actual experience of mobile phones is not very useful because our mobile phones rarely run to the limit. Compared with the starting score, I think power consumption and heat are more important. We not only dried the battery , but also took out the whole motherboard for the convenience of wiring. After all, it is an energy efficiency test. We must first ensure that the frequency is not reduced before we can measure the accurate power consumption. One thing to say is that the original heat dissipation of Mate60 Pro is also quite against the sky. I will go to the huge vapor chamber under the screen. It is really a treatment that I have never seen on a Huawei mobile phone. There are signs that this new Kirin should not be. It’s so easy to suppress it. After connecting the fake power and testing, it’s time to take out the golden signboard of our Geek Bay—the multi-core energy efficiency curve. Compared with other mobile phone CPUs, how is the energy efficiency of the Kirin 9000S? Hey, the overall energy efficiency of its CPU is slightly better than that of the Snapdragon 888 , especially the energy efficiency of the mid-to-high frequency part is stronger than that of the 888 and 8Gen1. However, the limit power consumption of the Kirin 9000S is relatively high. The price of breaking through 4000 points is the motherboard power of nearly 13W. And there is a saying that its energy efficiency is not as good as that of Snapdragon 865 and Kirin 9000 It is even farther away from the contemporary flagship 8Gen2. Of course, this is expected to be honest. After all, it uses a not-so-advanced technology and the self-developed core from the server side that is first installed on a mobile phone can achieve the current level. To be honest, it’s better than I expected. Speaking of which, the Kirin 9000S is also the first time that a mobile phone is equipped with hyper-threading technology. I saw that the audience has different opinions on this . What is the efficiency of the thread? I decided to test it. This is a quad-core and four-thread. This is a four-core and eight-thread. After hyper-threading is enabled, the performance has increased by 21.8%, and the power consumption has increased by 24.2%. From the perspective of energy efficiency, I think hyper-threading is no problem. The power consumption is quite normal , but I think the biggest problem with hyper-threading is that it is not necessary. The mobile phone applications are not designed for hyper-threading at all. If you want to make good use of hyper-threading, you must optimize them one by one. It should be said that it is not small . This budget is not as good as adding a few more cores to solve the compatibility problem , so why does Huawei still use hyperthreading? I think it’s more of an expedient measure, not to say that hyper-threading is so powerful that Huawei only uses it, but Huawei’s solution just supports hyper-threading. The Taishan core comes from the server. If you develop a completely new one without hyper-threading The architecture is dedicated to mobile phones, which is not good for cost control or delivery time. It is better to integrate the existing resources at hand and make a big move on the Taishan core modified from the server to meet the needs of the mobile phone . In this process, hyper-threading is still retained. The above is just my guess right or wrong, but one thing is that the efficiency of the Kirin 9000S in various applications may depend more on optimization. In addition to the CPU, this self-developed Ma Liang GPU is also our concern. The focus, however, in terms of GPU, we measured that the performance of Kirin 9000S is not as good as that of CPU. Its GPU peak performance is weaker than 888, and the energy efficiency of GPU is close to the level of 888. Compared with Kirin 9000 and 8Gen1 GPUs, it is worse. In fact, achieving the 888 level itself is not a shameful thing, but from the previous architecture analysis, this Ma Liang 910 GPU is still quite large , but it is still not smaller than the previous one and uses the old Adreno 600 series architecture. The 888 makes me a little bit disappointed, but it’s normal when I think about it. The self-developed CPU has been practiced by Huawei for a while. Kunpeng still has a lot of experience, but the self-developed GPU is only the first generation, right? In terms of efficiency, it is comparable to Qualcomm and ARM. The gap is also expected. If we compare it with 8Gen2, the gap is very large. On the bright side, its GPU energy efficiency is better than 865. In addition, the performance of the Kirin 9000S GPU under Vulkan is not as good as that of OpenGL. It is completely opposite to the current Qualcomm and MediaTek platforms. Through the energy efficiency tests of CPU and GPU, I believe that everyone has a relatively accurate understanding of the Kirin 9000S. When you first hear the product, do you feel that the tiger’s body is shocked, how can it be the same as Huolong 888? Hey, actually, let me tell you that the energy efficiency of the first generation of 888 is already in line with expectations. At this time, just think about Google’s Tensor and Samsung’s Exynos, and you will feel balanced. In addition, the energy efficiency curve does not explain all the problems. Let’s measure it . Huawei also has advantages in some places. For example, when we put the desktop on the desktop, the power consumption of the Mate60 Pro’s main board without the screen is only about 0.2W . This is a very low power consumption . The battery can be reduced to a minimum . During the test, we also found that the waveform of the Mate60 Pro is stable and there are very few small spikes. The application test has been done, right ? Isn’t it time to start battery life test and game test? It's a pity that the release of Mate60 Pro is so sudden that many game manufacturers and application developers haven't had time to adapt it to you. Like the game, we tried it. The picture quality of the king is currently only available in high-definition 60 frames. We played two games that were able to run normally with high quality. One is Nishuihan and the other is Honkai 3 Nishuihan mobile game. A while ago, the AI dialogue function was added to NPCs and it attracted many players. I also like this. I have heard about this game for a long time. In fact, this game also has certain requirements for hardware. You can see it when you look at the more beautiful scenes of Jinghu and Xianjuyuan. The picture is really good. The most stressful scene is still in Bianjing Hongqiao, which is basically the area with the most players in the game. There are all kinds of explosive players here. It will put a lot of challenges on the CPU. We set the maximum number of people on the same screen in the settings. After 40 people turned on the high frame rate, Mate60 Pro ran an average of 52 frames, slightly weaker than Mate40 Pro , and the power consumption of the whole machine is also higher. It can run to nearly 7W, which is quite scary. Although we ran against the water, we did not encounter graphics bugs . The frame rate is still quite smooth, but I think there is still room for adaptation and optimization. The good news is that even though the power consumption is not low , thanks to the excellent heat dissipation design of the Mate60 Pro body, we didn’t feel it when playing against the water. It is very hot. The highest temperature measured is less than 46 degrees. Its vc soaking plate distributes the heat evenly to the entire body. It should be said that the heat dissipation is quite good for the power consumption of the whole machine at 7W. The other one has not encountered any problems. My game is Honkai 3 and it can run smoothly on Mate60 Pro. After we played it here, Honkai Book 2 must be quite stressful, but what I didn’t expect was the performance of Mate60 Pro in Honkai 3 and The power consumption performance is actually better than that of Mate40 Pro. In fact, if you check the scheduling, you will also find that although the game has not been adapted to the hardware, the control strategy of Mate60 Pro here is quite real-time. CPU/GPU resources are basically allocated on demand. Fine, this is one of the advantages of the Kirin platform , that is, the architecture/interface is my own, and I have the authority to fine-tune every scene. I have a certain optimization potential. We will do it for you when the subsequent software and games work normally. More detailed battery life tests and more game tests to see how well Huawei has optimized this time. The above is our preliminary analysis of the Kirin 9000S. From a practical point of view, the Kirin 9000S is definitely not like the current flagship in the market. The platform is so strong and the energy efficiency gap is still relatively far. For those who care about cost performance, I may not rush to recommend you to choose Mate60 Pro, but this product itself is a milestone. It can be regarded as a real self-developed CPU GPU architecture. + Domestic technology and various objective factors are considered. With limited resources, Huawei has already made a good start. If you have the idea of supporting the domestic semiconductor industry, Kirin 9000S is at least a usable platform with great potential. I think we should give some applause and some encouragement. Some data about Kirin 9000S, including energy efficiency curves, have been uploaded to socpk.com. Interested students can go and see for themselves. Later, we will do more detailed tests on Mate60 Pro. So you must pay attention to our channel Geek Bay. There is a saying that I haven’t slept much for a few days , and I spent a lot of energy to get out of this video. If you think our efforts are worthwhile, you must press and hold Like one button three times to support us or come to the store in Geek Bay to buy some peripherals , then we will see you next time bye
Once again invaluable analysis from Geekerwan
The chip being made on a DUV 7 nm and being mostly comparable to the sd888 is pretty good honestly
Wow, the auto translate captions in YouTube are really good
This is actually pretty significant
There is an error in that video. On his website socpkl, he updated the results and the CPU is actually overall less efficient than 8 Gen 1.
China is catching up and wont slow down. The war of computation has begun
Off topic: is Youtube available in China, or is this Youtuber from HK? I believe Youtube is available in HK, but not in China.