Breaking through the network performance limit again! New H3C launches 1.6T intelligent computing switch

According to IDC’s forecast, the market size of the domestic AI field will continue to maintain a high-speed growth rate of over 20% per year in the next two to three years. In the scenario of intelligent computing centers, efficient integration of computing power resources and optimization of network transmission performance have become the key to the large-scale application of intelligent computing resources, with low latency and high bandwidth becoming standard configurations for intelligent computing networks. Over the years, Ethernet technology and new protocol standards have gradually developed, and the data transmission rate of each generation of Ethernet has doubled. Transitioning from 200G to 400G, and now gradually commercializing 800G, it can be confirmed that the trend towards 1.6T data center networks is inevitable.

In response to the surge in bandwidth demand in the era of intelligent computing, the H3C S9825-8C-G intelligent computing switch launched by New H3C supports 1.6T card configuration first, and the whole machine can support up to 16 1.6T OSFP interfaces, meeting the future demand for higher speed intelligent computing networks. From 400G to 1.6T, the signal output has increased by 4 times, which puts higher requirements on hardware wiring layout, line insertion loss, and signal integrity. The H3C S9825-8C-G single board adopts advanced 1.6T signal connectors and high-frequency boards in China. Through the fine design of single board stacking and wiring channels, the high quality of 112G PAM4 signal output is ensured. At the same time, strict simulation and optimization design were carried out on the chips, connectors, and PCBs in the link, and the overall link loss of the optical outlet was controlled within 5dB, meeting the standard of line insertion loss and achieving industry-leading performance levels.

Higher port speeds not only increase the complexity of signal design, but also pose higher requirements for heat dissipation design. H3C S9825-8C-G adopts thermal simulation technology for heat dissipation simulation, front and rear straight air duct design, and optimized PCB board wiring layout, which meets the power consumption requirements of mainstream 1.6T modules in the market. The product is also equipped with a high-performance fan with 3+1 redundancy design, supporting silent energy-saving mode, and has the ability to deploy a mixture of liquid and air cooling, which can significantly reduce system power consumption and improve operational efficiency.

In addition, the product is embedded with NAI intelligent native technology, combined with AI ECN and AI traffic profiling technology, to monitor the transmission delay, queue length, and cache changes of different queues in real time, automatically adjust the transmission threshold, and achieve end-to-end lossless network. In the face of network congestion issues, the product has microsecond level fault self-healing capability for DPSH data plane, which can sense faults and actively switch detours, ensuring that upper layer applications have no perception and AI training is not interrupted.

In the field of intelligent computing networks, New H3C Group adheres to a standardized Ethernet technology stack, maintains open and compatible architecture, achieves optimal allocation of networks and diverse computing power systems, and has a leading full range of 200G/400G/800G high-performance switch products in the industry, which can meet the flexible networking needs of different users.