Just now, Huawei officially released a new generation of Kirin processors under HiSilicon in Germany-Kirin 990. This is the world's first 5G SoC based on 7nm and EUV process (Extreme Ultraviolet Lithography). It is worth mentioning that Kirin 990 will integrate the 5G baseband chip Balong 5000, which can realize 5G network without external 5G chip, and support SA/NSA two 5G networking modes.
In terms of CPU, Kirin 990 uses 4 large cores + 4 small cores, which are 4 A76 large cores and 4 A55 small cores. For GPU, Kirin 990 uses ARM Mali G76 with a total of 16 cores. For NPU, Kirin 990 uses a self-developed Da Vinci architecture NPU, composed of dual core Ascend D110 Lite and Ascend D100 Tiny.
According to Huawei introduced, Kirin 990 5G is the number of a transistor more than 10 billion of the world's first mobile terminal chips , reaching 10.3 billion transistors, compared with the previous Kirin 980 increased 4.4 billion transistors. Huawei’s consumer business CEO Yu Chengdong said that Huawei’s flagship new machine Mate 30 will meet with consumers in Munich on September 19, and it is expected that Mate 30 is expected to be the first model equipped with Kirin 990.
Significantly improved energy efficiency ratio
Different from the 4G SoC+5G Modem mode adopted by the industry, Kirin 990 5G adopts the industry's most advanced 7nm+ EUV process and integrates 5G Modem on the SoC for the first time. The board-level area is 36% smaller than other solutions in the industry. According to reports, the biggest feature of Kirin 990 5G is that the energy efficiency ratio has been greatly improved.
For 5G, the power consumption ratio has always been an urgent problem to be solved. Kirin 990 5G supports ultra-high 5G rates, achieving 2.3Gbps peak download rate and 1.25Gbps uplink peak rate in the Sub-6GHz frequency band. It is the first to support 5G dual cards. While one card can use 5G to surf the Internet, the other card can receive VoLTE high-definition voice calls, realizing the best 5G dual card experience in the industry.
In addition, Kirin 990 5G achieves the best 5G energy efficiency in the industry. Compared with the traditional 4G SoC+5G Modem solution, the power consumption performance is 20% better, bringing a longer-lasting 5G experience. If it is on a moving high-speed rail, it can also be adjusted based on machine learning according to the speed of travel, so as to reach the optimal solution.
At the 2019 Mobile World Congress (MWC) held this year, China Mobile, China Telecom and China Unicom all announced that 5G networks will adopt independent networking as the target architecture. In the two architectures of the 5G network: Standalone (SA) and Non-Standalone (NSA), the 5G carrier of the NSA architecture only carries user data, and its control signaling is still transmitted through the 4G network , Its deployment can be seen as adding new carriers to the existing 4G network for capacity expansion, and the SA architecture is an architecture that adopts new design ideas. While introducing new network elements and interfaces, it also adopts network virtualization and software-defined networks on a large scale. And other new technologies, combined with the new 5G NR air interface technology, the technical challenges faced by its protocol development, network planning and deployment, and interoperability will surpass those of 3G and 4G systems.
However, not all of the 5G chips currently announced can support the SA networking architecture. Huawei said that Kirin 990 5G is not only a chip designed to meet current needs, but also a chip designed for the future, supporting two major networking architectures. The most important thing is that, regardless of heavy load or light load, the performance can be as much as 8 times better than other chips.
AI performance is upgraded
According to Yu Chengdong, the Kirin 990 5G chip has outstanding AI performance. The AI performance has increased by 12 times in two years. It not only provides the best AI computing power, but also provides the most optimized energy efficiency. Kirin 990 5G also supports HiAI, TensorFlow, Android NN and other models and frameworks. Yu Chengdong said that whether it is a heavy-load or light-load model, 990 5G can provide better energy consumption. The Kirin 990 5G chip also supports more than 300 operators and 90% of visual computing neural networks.
In addition, Kirin 990 5G introduces a new type of AI architecture "big core + micro core". Yu Chengdong gives an example: you can use trucks to load goods as an analogy. If you use heavy trucks to load a small amount of goods, this is naturally a waste of resources. The "big core + micro core" architecture is to allow heavy trucks to load heavy goods, and light trucks to load small weight goods. Not only will it not cause performance waste, but it can also improve the efficiency of the entire structure and achieve optimal energy efficiency.
The Kirin 990 5G chip also uses a CPU architecture of 2 large cores + 2 medium cores + 4 small cores. Compared with the Snapdragon 855, the large nuclear energy efficiency can be improved by 12%, the medium nuclear energy efficiency is improved by 35%, and the small nuclear energy efficiency is improved by 15%.
Kirin 990 5G also upgraded the GPU to 16-core GPU Mali-G76. Compared with the Snapdragon 855, the graphics performance is improved by 6% and the energy efficiency is improved by 20%.
Concluding remarks
Since 2009, it has been committed to 5G research and development. Huawei now has end-to-end solutions that support SA and NSA from 5G core network to base station, to 5G mobile phones, 5G CPE and 5G Modem. The Kirin 990 5G released this time is also the industry's first full Netcom 5G SoC, which simultaneously supports SA/NSA 5G dual-module networks, supports TDD/FDD full frequency bands, and is suitable for all 5G network frequency band requirements.
As an explorer in the 5G and AI era, it is hoped that Huawei can give full play to the technological advantages of the Kirin platform in the future, fully open its chip capabilities to developers, and allow more developers to join the development queue of all-scenario and intelligent new applications.