Yunmai Xinlian released the first domestic self-developed smart network card product supporting 2-port 100G RDMA

With the continuous deepening of my country’s new digital infrastructure construction, the rapid development of artificial intelligence, big data, and industrial Internet applications, the demand for data processing has grown rapidly. According to IDC data, the total amount of global data is expected to reach 175ZB in 2025, and the amount of data owned by China will reach 48.6ZB, ranking first in the world. In terms of growth rate, the average growth rate of China’s data volume and data center traffic is faster than that of the world, and it is still in the stage of rapid development.

Faced with the increasing pressure of massive data transmission and processing, the current 10G and 25G networks can no longer meet the needs of modern intelligent computing, and 100G or even 400G data center networks have gradually become the development trend of the new generation of digital infrastructure and the interconnection of a new generation of heterogeneous computing power. main network infrastructure.

On January 5th, China’s leading provider of digital infrastructure base DPU chips and solutions, Yunmai Xinlian officially released the smart network card (DPU) product metaFusion-200 for 100G bandwidth. This product is the first 100G DPU product based on the FPGA platform in China, which integrates OvS virtualization network offloading function and RDMA storage network function, and supports cloud computing such as elastic bare metal, elastic cloud host, elastic container cloud, and smart computing cloud. Application scenarios can meet the business demands of cluster interconnection and computing power expansion in core scenarios such as general computing power, high-performance computing, AI, and cloud storage, and can help users achieve stable and smooth upgrades of cloud data center networks at lower costs and higher efficiency , to accelerate the development of infrastructure towards high efficiency and low carbon.

Yunmai Xinlian metaFusion-200

At the same time, in order to meet the diverse business needs of users, Yunmai Xinlian simultaneously launched the smart network card product metaConnect-200 that supports 2x100G network ports, providing RoCEv2 lossless network customization capabilities and virtual network offload acceleration capabilities, mainly used in high-performance computing , high-performance storage and other scenarios, which can significantly improve the deployment and operation efficiency of users’ high-performance networks. Test data shows that metaConnect-200 can effectively improve network bandwidth utilization by 18% compared with mainstream smart network card products that support RoCEv2 on the market.

Yunmai Xinlian metaConnect-200

The newly released metaFusion-200 and metaConnect-200 products are based on the core technology R&D and design of Yunmai Xinlian’s independent intellectual property rights, support self-developed HyperDirect technology, can effectively accelerate the computing efficiency of GPU and AI chips, and have a good relationship with domestic mainstream GPGPU Compatibility. At the same time, the self-developed programmable congestion control algorithm platform can also help users design and apply suitable congestion control algorithms according to different business types, and improve the overall end-to-end throughput and reliability of the network. It is reported that based on good cooperation and communication with users and excellent solution capabilities, this series of products has been welcomed by the market as soon as it is launched, and has now received commercial orders.

The competition between artificial intelligence and big data has entered the era of low-level computing power competition. As Moore’s Law slows down and CPU performance growth encounters bottlenecks, smart network cards (DPUs) have become a new computing power growth engine. Yunmai Xinlian smart network card (DPU) opens up a new growth space for the improvement of data center cluster computing power by offloading, releasing and expanding the computing power of the host. Use DPU to improve network performance, achieve high-throughput and low-latency performance in long-distance, large-scale transmission scenarios, truly accelerate the balanced distribution of computing power across the country, improve the utilization efficiency of computing power in the west, and accelerate the construction of an integrated big data center , Contribute to the implementation of “East Counting and West Counting”.

Since its establishment in May last year, Yunmai Xinlian has been committed to the research and development of cloud data center network chips and technological innovation. Through independent research and development and innovation practices, it has focused on RDMA high-performance network transmission, network and storage infrastructure service offloading, and IO virtualization. The three major DPU core capabilities have achieved breakthroughs in the field of large-scale network congestion control, taking the lead in reaching the international leading level and filling the domestic gap. Yunmai Xinlian said that the company has focused on the independent research and development of DPU chips since its establishment, and will continue to invest in technological innovation and practice in the future to bring greater value to users and contribute to breakthroughs in the field of large chips in China.