By hmimcu 
The topic of cross-border core making has always been a hot topic in China. Especially for technology giants led by BAT, their achievements on the path of core making have attracted attention from the Chinese people.
Recently, Tencent Cloud announced on its official WeChat account that after successfully “lighting up” in March last year, its self-developed video encoding and decoding chip “Canghai” has been mass-produced and put into use tens of thousands of pieces!
In scenarios such as cloud games and direct on-demand, Canghai has currently provided services to Tencent’s self-developed business and public cloud customers.
In the latest released MSU * hardware video encoding competition results, Canghai won first place in all 8 ratings of the 2 tracks he participated in.
It is understood that the MSU hardware video encoding competition is the most influential top-level competition in the field of video compression. So far, it has been held for 17 consecutive sessions, attracting well-known enterprises at home and abroad, including Amazon, Nvidia, Intel, AMD, Byte, Kwai, Ali and Tencent.
image.png
From various indicators, at the same bit rate, compared to industry standard hardware such as GPUs, Canghai can significantly improve picture quality; In the high frame rate range of 120fps, under the same quality, Canghai further saves more than 10% bit rate compared to the industry leading level; In addition, the encoding time for a single frame of 1080p in Canghai chip is only 4 milliseconds, and the throughput performance is almost unaffected by encoding parameters and video types.
This also means higher definition with the same bandwidth, more bandwidth efficient with the same image quality, and faster encoding.
According to Tencent Cloud, the “Canghai” chip focuses on solving the impossible triangle of high image quality, low latency, and low cost in video encoding and decoding.
image.png
In terms of algorithm, the “Canghai” chip fully implements all mainstream encoding tools such as high-precision motion search, full rate distortion optimization, and efficient adaptive quantization, and integrates leading technologies in Tencent Cloud software encoder rate control. The compression rate is more than 35% higher than the market standard.
In terms of architecture, further offloading the video pre analysis computing unit into hardware significantly reduces the dependence of tasks on CPU performance in high-density transcoding scenarios. By adopting a flexible multi-core extension architecture, high-performance encoding pipeline, and hierarchical memory layout, it meets the requirements of high throughput, low latency, and real-time performance. Tencent’s rich multimedia application scenarios provide sufficient analysis and validation conditions for Canghai.
New progress has been made in the implementation of “Canghai”/”Zixiao”/”Xuanling”
Although Tencent started from QQ, the well-known Tencent business also revolves around its software ecosystem. But many people may not be clear about Tencent’s layout in terms of chips.
As early as May 2018, Tencent Chairman and CEO Ma Huateng revealed some ideas about chips in a forum speech. He proposed that doing software and services seems a bit far from chips. But in fact, Tencent has built many data centers, including the cloud, which has a certain influence on the demand for many server ends, even chips. The application services are massive, and the user scale is particularly large, which can force the chip design industry to make targeted designs. If we could still intervene and support the development of some chips, it might be even better
Like many cross-border chip giants, Tencent initially chose to invest in this field. In August 2018, AI chip startup company Suiyuan Technology obtained a Pre A round of financing of 340 million yuan, led by Tencent.
At the 2021 Tencent Digital Ecology Conference, Tencent announced its important progress in chip development to the world: among the three self-developed chips, “Zixiao” is mainly used for AI, “Canghai” is mainly used for video, and “Xuanling” is mainly used for intelligent networks. Tencent has provided specialized explanations on the characteristics of these chips, and in summary, the performance of the three chips has significantly improved compared to similar products.
The “Canghai” chip has been introduced in detail in the previous section. It implements all mainstream encoding tools such as high-precision motion search, full rate distortion optimization, and efficient adaptive quantization in algorithm, and integrates leading technologies in Tencent Cloud software encoder rate control. It can achieve real-time video encoding of 1080P 60Hz, with a compression rate increase of over 30% compared to the industry. It may contribute to Tencent’s success in WeChat QQ and other social communication software play an important role in video calling and gaming entertainment. In the current era where users are increasingly picky about video image quality, the “Canghai” chip is bound to have great potential in the video field!
Secondly, it is the “Zixiao” chip. The most remarkable feature of this AI reasoning chip is that it combines picture and video processing, natural language processing, search recommendation and other scenarios, and its performance is 100% higher than that of the industry. The chip architecture is optimized by adding a computer vision CV accelerator and a video codec accelerator inside the chip.
Although Tencent did not explicitly specify its application scenarios at the time, it is not difficult to speculate that it may combine Tencent’s related video and entertainment businesses, such as Tencent Video, Tencent Sports, Tiger Teeth, Douyu, and other video platforms. For example, when searching video content through voice, it will further improve search speed.
The “Xuanling” smart network card chip is positioned to accelerate the performance of cloud hosts. By optimizing the chip architecture in scenarios such as CVM/BM/container, the virtualization, network/storage IO, and other functions that originally ran on the main CPU are moved down to the chip, achieving zero occupancy of the main CPU and improving the performance by four times compared to industry products. The goal of this chip may be related to cloud computing, and further, it may be related to cloud gaming. The gaming business is crucial for Tencent, while cloud gaming is oriented towards the future. Through the Xuanling smart network card chip, Tencent may further complete its in-depth layout in the field of cloud gaming
For Tencent, cross-border core manufacturing is undoubtedly a difficult path. Especially for self-developed chips, there is a huge initial investment, slow returns, and a long way to go in research and development. But at the same time, it can also be seen that the birth of the three major specialized chips, Canghai, Zixiao, and Xuanling, has also brought tremendous assistance to Tencent in solving some of its own special business pain points. By developing their own customized chips, not only can they meet their business needs and improve efficiency, but they can also make faster layouts for future new technologies.