This Wednesday,Xilinx has launched a real-time server reference architecture for video transcoding, as well as a new accelerator card alveo u30.
U30, as the latest product of Syrith's Alveo series of accelerator cards, focuses on achieving high channel density, which is also proposed by Syrith
According to sarinx, with sarinx's new real-time server reference architecture, suppliers can minimize costs while delivering high-quality live video services. This means that saisiling's data center ecosystem will further expand and layout.
Two-pronged approach to solving the problem of webcast
In 2020, under the pressure of the new crown epidemic, video live broadcast ushered in rapid growth. According to the Internet industry report provided by mhojhos research, the global market value of real-time video streaming will reach 31 billion US dollars in 2020, and it is expected to reach 94 billion US dollars by 2026.
The rising market value of video streaming means that the number of users of live video is increasing rapidly and the demand for video quality is higher. As a result, the cost of bandwidth for live video which occupies a large amount of network traffic is higher.
In the face of these two problems, at this conference, based on the real-time server integrated machine reference architecture and accelerator card alveo u30, Xilinx launched two sets of solutions respectively.
1. Optimize cost per channel based on u30
Facing the webcastThe all-in-one machine is also suitable for safe city, intelligent retail, E-sports and other application scenarios.
The new accelerator card u30 introduced this time adopts the half height and half length shape and single slot design, and supports two encoding formats, h.264/avc and h.265/hevc. Each card can real-time 2x4kp60 UHD transcoding, and can support up to 48 channels. In addition, u30 supports transcoding and decoding with low delay and ultra-low delay, which can reduce the delay to 100ms while ensuring the video quality. In terms of power consumption, u30 provides a low-power design scheme of less than 40W, and the maximum power consumption is limited to 75W.
Compared with similar competitors, the high-density u30 solution has its own advantages. For example, its video quality is no less than NVIDIA T4, and it can provide a higher density than T4. In terms of system power consumption, it is less than 20% of T4.
According to the official data of Xilinx, if the Xilinx RT server is compared with the HPE ProLiant DL380 server, the performance of a Xilinx RT server equipped with 8 alveo u30 accelerators is equivalent to that of 4 HPE ProLiant DL380 servers equipped with 32 NVIDIA T4 accelerators, and the former has 4 times the advantage of throughput per card, and the hardware cost is reduced by 6 Power consumption cost is reduced by 5 times. In addition, the u30 can also speed up Intel's servers, according to the company.
2. Optimize cost per bit based on U50
Face the FaceThis solution can reduce the bit rate and repeatability cost, and minimize the cost of each stream under the premise that the cost of video per Gigabyte remains the same. It is suitable for scenes with high video quality requirements.
The U50 Accelerator Card, launched in August last year, is the industry's first lightweight PCIe Gen4 adaptive computing acceleration card and faces all servers, various cloud and edge data center applications, including network and storage acceleration. U50 use the Selling UltraScale architecture, the first use of half-high half-length shape size and less than 75 watts of low envelope power consumption. the card supports high-bandwidth memory (HBM2) and 100 Gbps network connections and supports fourth-generation PCIe and CCIX interconnection standards.
According to sarinx, the U50 can support hevc of 1080p and 120. If other equivalent software infrastructure is used to achieve the same performance, it needs 5 HPE ProLiant DL380 servers plus 10 very expensive and most powerful platinum level devices. But if you use the U50 solution, you only need one HPE ProLiant DL385 server to build eight accelerator cards of alveo U50. As a result, the throughput of each node of Xilinx solution is 5 times, the hardware cost can be reduced by 6 times, and the power consumption can be reduced by 3 times.
Additional software solutions without FPGA development experience
For the two integrated server solutions, Xilinx also gives a relatively simple and convenient software solution.
In the server optimization software solution stack of Xilinx, because of the cooperation with AMD, the epyc processor of AMD is mainly used. At the bottom is the accelerator card of Xilinx alveo U50 or u30. The upper layer of the accelerator card is the binary file of Xilinx accelerator, which mainly supports the functions of encoding, decoding and video processing. At the top of the file layer are Xilinx media acceleration API and runtime API, supporting higher-level applications such as system layer and software layer. The top layer is ffmpeg command line calculation framework.
Server management is mainly done in the server, using resource management or XRM technology. In addition, some other multi server management and multi stack management are realized through kubernettes management function.This means that with the solution of sarinx, only a few characters need to be changed to realize more efficient video transcoding without FPGA experience.
In addition, as sarinx cooperates with wowza, wowza's GUI and video live engine are also integrated into sarinx's real-time server reference architecture.
Xilinx said that the reason for providing additional solutions in software solutions is to provide customers with a very comprehensive delivery plan to support live video.
Continuously layout data center ecosystem and develop incremental market
Xilinx's layout in the video field has lasted for ten years. It cooperates with OEMs in the field of live video for a long time. It also has many years of experience in vertical fields such as industry, medicine, automobile, etc. at present, its focus is to support the workload of data center and enhance the algorithm in the field of coding and decoding.
In the current real-time video transcoding integrated machine market, Intel, NVIDIA, Broadcom and other enterprises occupy half of the market, while Xilinx occupies a relatively small market space. The reason why the integrated machine is launched is that we hope to provide customers with replicable experience and help customers to achieve faster deployment.
Compared with other competitors, the total cost of ownership of the all-in-one machine provided by sarinx is lower, which shows that sarinx hopes to gain more customers' favor through lower cost, so as to establish a more complete ecosystem.
According to Zhong Yi, senior sales director of data center in Greater China, Xilinx has always focused on the continuous expansion of data center, especially ecosystem. In the last fiscal year, more than 10000 enterprises and academic units were trained by FY20, nearly 1000 partners joined alveo, and more than 130 applications based on alveo were released.
For sarinx, we currently have FPGA, SOC, heterogeneous MPSoC and 3dic, and ACAP, which is dedicated to data explosion and Moore's law failure. Since October 2018, a number of alveo series accelerator cards suitable for different application scenarios have been launched to continuously inject new vitality into the data center.
This time, sarinx aims to speed up the live video broadcasting. What's the next time?