On April 17, Tencent Cloud announced that its self-developed video transcoding chip named Canghai had won all eight top scores in two categories at the latest MSU Video Codecs Comparison held by Moscow State University after months of rigorous testing.
In early 2019, Tencent launched the development of the Canghai chip. The firm conducted in-depth research on hardware encoding and decoding algorithms, architectures, and engineering. By allocating different computing units, they were able to stack a large number of processing tasks. It incorporated dedicated functional modules for Canghai chips to enable built-in video encoding acceleration, reducing redundant calculations and branch judgments. This resulted in lower power consumption and cost.
In March of last year, Canghai chip successfully passed tests and is now in mass production with tens of thousands of units deployed. It provides services for Tencent‘s self-developed businesses and public cloud customers in scenarios such as cloud gaming and live streaming.
Canghai chips can be beneficial for live streaming platforms with less than 200 viewers as they help reduce computing costs. Additionally, in scenes with 4K and 8K ultra high definition and high frame rate, Canghai chips provide superior transcoding quality and low latency compared to software encoding.
The Yaochi graphics card of Canghai chip can adapt to various vendor GPUs, making it suitable for cloud gaming. Additionally, the Canghai chip offers a high-density transcoding capacity of 1024 channels per machine, which reduces the cost of video transcoding.
Chips are the most essential part of the hardware. With the joint optimization of software and hardware, Tencent Cloud has also gained certain product advantages in media processing (MPS). Now, Tencent Cloud’s services provided to customers include audio and video quality enhancement, cloud-native recording, real-time special effects rendering, intelligent editing, and content analysis.
For scenarios with strong business needs, Tencent has a long-term plan and investment in chip development and will combine computing power scenarios to promote the scale implementation of self-developed chips. Currently, Tencent has achieved end-to-end design and verification coverage of chips.
Apart from the Canghai chip, Tencent Cloud also disclosed that its self-developed AI inference chip Zixiao has been put into use in internal operations and can provide up to 3 times computing acceleration performance and over 45% overall cost savings.