1, Kirin 990 4G
Since the 4G version of this generation is actually the "enhanced version (the process technology has not been upgraded)" of the previous generation Kirin 980, it is necessary to review the various advantages of the previous generation architecture inherited by this generation (the architecture of the CPU and GPU has not changed).
First of all, the CPU of the Kirin 990 still adopts a three-level energy efficiency architecture called "Kirin CPU Subsystem" - consisting of two A76 super large cores (main frequency 2.86Ghz), two A76 large cores (main frequency 2.09Ghz) and four A55 small cores (main frequency 1.86Ghz).
In addition, the previous generation of intelligent scheduling mechanism called "Flex-Scheduling" also has an inheritance, which can not only flexibly call each core, but also predict the peak of traffic required by the mobile phone through its powerful AI performance to avoid lag.
In other words, compared to the previous generation, the main frequency of all eight cores of the CPU has been increased - so the single-core performance has increased by 11%, and the multi-core performance has increased by 22%. In addition, this time, Huawei has begun to "magically modify" the public version of the architecture, so the core energy efficiency has been improved.
According to the official statement, the two large cores have the largest improvement in energy efficiency, followed by four small cores, and the two super large cores have the smallest improvement in energy efficiency. Therefore, this time in terms of single-core power consumption, it is slightly higher than the previous generation.
This is why, in the case of the ARM public CPU architecture and process technology, the power consumption of this generation can be controlled so well. Therefore, the use of public version architecture is only a premise, as long as the design is good and the actual experience is good, it is the performance of "cattle"!
As for "AI performance", this generation of Huawei's self-developed da Vinci NPU architecture can be well said. Specifically, this architecture uses 3D Cube to accelerate matrix operations, thereby greatly increasing AI computing power per unit of power consumption.
At the same time, in order to improve the completeness of AI computing and the computing efficiency of different scenarios, the da Vinci architecture also integrates a variety of computing units such as vectors, scalars, and hardware accelerators. It supports multiple precision calculations, supports data accuracy requirements in both training and inference scenarios, and realizes the full-scenario requirement coverage of AI.
In addition, this time it adopts an innovative "large core + small core" NPU architecture, in which the small core is designed to reduce power consumption - it is used to carry computing tasks such as intelligent voice wake-up, voice recognition, and face unlocking, and can also wake up large cores and participate in co-computing.
Finally, on the GPU side, although the Mali-G76 architecture is used, the number of cores has increased from the previous generation of ten cores to sixteen cores, and the frequency has been reduced from the previous generation of 720Mhz to 600Mhz - it can be seen that the "large-scale low frequency" route is taken.
In addition, this time Huawei has added a smart cache device called "Smart Cache" to the GPU, which can play the role of intelligent data distribution and relieve the pressure of internal memory DDR, which is to put it - power consumption is reduced by 12%!
Coupled with the reduction in power consumption caused by the reduction of core frequency, this time the GPU not only achieved a significant increase in performance (more than 30%), but also slightly reduced power consumption! Good guy, with the process technology and architecture all unchanged, Huawei has made a flower!
2, Kirin 990 5G
Finally, the updated "protagonist" of this generation is about to play. First of all, it is not only the world's first flagship chip that integrates 5G baseband, but also the world's first flagship chip with the number of transistors exceeding the 10 billion mark.
At the same time, this is also the industry's first 5G all-Netcom SoC (and supports SA and NSA) and the first TSMC 7nm+EUV process chip (this process has significantly improved energy efficiency); In addition, in terms of 5G speed and 5G energy efficiency, records have also been broken.
In addition, because its NPU is based on the "dual large core + single small core" design of the da Vinci architecture, it ranks first in the world in terms of AI computing power (ahead of the Apple A13 released later). Even if the Snapdragon 865 was released later, it failed to shake the Kirin in the slightest.
In fact, this generation has also made progress in ISP, thanks to the new ISP 5.0, the throughput rate has been increased by 15%, and the energy efficiency has also increased by 15%, and the processing power of video shooting has been fully optimized.
In addition, this generation of ISPs also launched the world's first "dual-domain joint video noise reduction technology" - video noise reduction ability increased by 20%, which means that shooting videos in low-light environments will be clearer. Then there is real-time video post-processing rendering technology based on AI segmentation, so that mobile phone videos can also have the texture of movie color grading.
In addition, it also applied the SLR-level "BM3D" technology to the mobile phone for the first time, thereby increasing the noise reduction ability of photos by 30%! And based on ISP capability upgrades and surging AI computing power support, this generation also supports cool "Face AR".
Since the Kirin 990 5G is only the main frequency of large cores and small cores in terms of CPU architecture, there is no change in the single-core running score of the performance test, and there is not much difference in multi-core performance.
As for the GPU part, it cannot be said to be similar, it can only be said to be exactly the same; Of course, thanks to the upgrade of the process technology, GPU power consumption still has a slight advantage. Similarly, the power consumption of the CPU part is also better. Coupled with the power consumption advantages generated by the integration of 5G baseband, this generation is the legendary "ice unicorn"!
So, why does the Kirin 990 5G not pay attention to the improvement of performance when the process technology is advantageous? The reason is the integration of 5G baseband - most of the area saved by process improvement is used to stack baseband.
3. Performance comparison
Here is to invite the Kirin 990 5G to fight, and now some mainstream mid-range chips to compare, to see how the Kirin combat power is after three years. First of all, a particularly difficult one - Qualcomm's latest "mid-range God U" Snapdragon 7 + Gen2.
In terms of CPU single-core performance, Snapdragon has an advantage of more than 50% (after all, the ultra-large core is X2 architecture), while the single-core power consumption has only increased by 20% (after all, the 4nm process technology); The gap in multi-core performance narrowed to 27%, and Kirin performed well in this regard.
However, when it comes to the GPU link, the gap is relatively large, and Kirin is directly behind Snapdragon by more than 80%! The power consumption is expressed between the two. Of course, the unicorn's side just lost to time, and this comparison result is not surprising.
Next, please invite the Snapdragon 870, which is forced to become a "Contrast God U", to appear. In terms of CPU single-core performance, the 870 leads by 32%, while the power consumption is also 20%; In terms of multi-core performance, Kirin lags slightly behind by 9%, and it can be said that the gap is not obvious.
In terms of GPU performance, Kirin lags behind by 32%, but the power consumption performance has a 35% advantage over the 870! It can be seen that Huawei did put a lot of hard work into the GPU energy efficiency of the Kirin chip back then - this is also the key reason why there are so many "nail households" of the Kirin 990 series!
The Qualcomm camp finally compared the "old mid-range God U" Snapdragon 778G, the CPU performance is comparable, and the Kirin is significantly ahead of 33% in GPU performance! Although the overall power consumption performance of 778G is greatly superior, it is all the credit of the architecture and process.
Since MediaTek's Dimensity 1080 can't even beat the Kirin 980, there will be no comparison here, and the following is still compared with the A13 of the fruit. The process process of this A13 is N7P like the Snapdragon 870, and Kirin has a little advantage in this regard.
Although the CPU single-core performance fruit is 1.69 times that of Kirin , its power consumption is a full double that of Kirin ! Although the Kirin process technology has a leading edge here, the gap is not so large, it can only be said that the fruit pile is too strong, and the power consumption also increases.
As for multi-core performance, the difference between the two is small, with Kirin only lagging behind by 13%. In terms of GPU performance, the fruit is 62% ahead, while the power consumption has also increased by 73%, and the energy efficiency performance is still not as good as that of Kirin . On the whole, even if you compare the pile of crazy devil fruit brother, it is not intimidated!
Summary:
The biggest advantage of the Kirin 990 series is energy efficiency, followed by AI performance. Because although its GPU performance has improved greatly, it is still not as good as the contemporaneous Snapdragon 865, not to mention the A13 of the pile mad fruit brother.
Therefore, in the Android camp, the Kirin 990 series focuses on AI performance, which can crush the crowd in this regard; On Apple's side, although the performance is not good, but the energy efficiency performance is more advantageous, as for 5G and signal, Huawei is properly crushing Apple!
Even today, in 2023, its energy efficiency performance is still enough, and the GPU performance is not weak; And some models equipped with Kirin 990 5G (such as P40 Pro/Pro+) also have a 90hz screen high brush, which can be used smoothly for a few years!
END
Committed to the exploration and collation of mobile phone knowledge