laitimes

Report the progress of Tencent's self-developed chip "Canghai"

After successfully "lit up" in March last year, our self-developed video codec chip "Canghai" has been mass-produced and put into use tens of thousands of pieces.

In scenarios such as cloud gaming and direct on-demand, Canghai has provided services to Tencent's self-developed business and public cloud customers.

In the newly announced results of the MSU* hardware video coding competition, Canghai won the first place in all 8 scores of the 2 tracks in which it participated.

* The most influential top competition in the field of video compression. So far, it has been held for 17 consecutive years, attracting well-known enterprises at home and abroad, including Amazon, NVIDIA, Intel, AMD, Byte, Kuaishou, Ali and Tencent.

Report the progress of Tencent's self-developed chip "Canghai"

From the perspective of various indicators, under the same bit rate, compared with standard hardware such as GPUs in the industry, Canghai can greatly improve the picture quality; In the high frame rate of 120fps, under the same quality, Canghai further saves more than 10% of the bitrate compared with the industry-leading level; In addition, the encoding time of a single frame 1080p of the Canghai chip is only 4 milliseconds, and the throughput performance is hardly affected by the encoding parameters and video type.

——The same bandwidth is more high-definition, the same picture quality is more bandwidth-saving, and the compilation is fast.

8 firsts, the "impossible triangle" of the sea

Canghai chip focuses on solving the impossible triangle of high image quality, low latency, and low cost in video encoding and decoding.

Report the progress of Tencent's self-developed chip "Canghai"

In terms of algorithm, Canghai chip has fully realized all mainstream coding tools such as high-precision motion search, full-rate distortion optimization, and efficient adaptive quantization, and integrates leading technologies such as Tencent Cloud software encoder bitrate control, and the compression rate is more than 35% higher than that of market standard products.

In terms of architecture, the video pre-analysis computing unit is further offloaded to the hardware, which greatly reduces the dependence on CPU performance of tasks in high-density transcoding scenarios. Through flexible multi-core expansion 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 verification conditions for Canghai.

The scene verification displays:

For cloud gaming scenarios, the Yaoguang board card of Canghai chip can flexibly match the GPU of different manufacturers;

For video transcoding scenarios, Canghai chip provides high-density transcoding capabilities of 1024 channels on a single machine, minimizing transcoding costs.

In 4K/8K ultra-high-definition and high-frame rate video scenarios, Canghai chip can achieve low-latency and high-quality transcoding that cannot be achieved by software encoding.

……

The first of 21 items, video cloud capabilities continue to lead the industry

As the cloud enterprise that understands audio and video the best, Tencent Cloud Audio and Video also provides customers with lower cost, lower latency, and more extreme compression processing effects through joint software and hardware tuning.

In this MSU Cloud Video Transcoding Competition, Tencent Cloud Media Processing Product MPS (Media Processing Service) also stood out among many international manufacturers, winning the first place in 21 evaluations of H264 and HEVC/AV1 tracks.

Among them, the TXAV1/V265 encoder ranked first in all objective and subjective indicators of PSNR, SSIM and VMAF in all resolution tests. At the same time, Tencent O264 encoder also won the first place in 6 indicators of the track.

Under the same transcoding cost, using Tencent Cloud MPS media processing products can save more than 20% of bandwidth and storage costs compared with other competitors. At present, Tencent Cloud media processing product MPS processes an average of more than 21 billion minutes of video per month, helping customers save hundreds of millions of dollars in bandwidth and storage costs every month.

In fact, this is not the first time that Tencent Cloud MPS has occupied a leading position in the industry evaluation. In a report released by SLC, a professional streaming media technology community, last year, Tencent Cloud MPS also surpassed many international vendors and won the best performance in all three major scenarios.

At present, Tencent Cloud Services has 90% of domestic audio and video customers, and is still increasing its overseas track, joining hands with all walks of life to move towards the era of true interconnection.

Facing scenarios with strong business needs, Tencent has long-term chip R&D planning and investment. At present, Tencent has achieved full coverage of end-to-end chip design and verification, and a number of chips have achieved large-scale landing.

The AI inference chip "Zixiao", which adopts self-developed memory and computing architecture and self-developed acceleration module, provides up to 3 times the computing acceleration performance and more than 45% overall cost savings, and has been put into use in internal business;

The high-performance network chip "Xuanling" adopts self-developed network, storage, and computing acceleration solutions to achieve "0" occupation of host CPU and up to 4 times the performance improvement, helping to build a next-generation high-performance network infrastructure.

In the future, we will continue to deepen the underlying software and hardware, explore the best performance and the best cost performance, and help more enterprises and developers innovate on the cloud."

The road is blocked and long, and the line is coming.

Read on