CFM has updated the domain of its english website from en.chinaflashmarket.com to www.memorymarket.com, Please be informed.

Sugon Launches ScaleFabric, China's First Self-Developed Native RDMA High-Speed ​​Network

By: QIN 1 day ago

On March 12th, Sugon officially launched its first self-developed 400G lossless high-speed network, ScaleFabric, in Zhengzhou. This is China's first InfiniBand-like network product based on a native RDMA architecture. From the underlying 112G SerDes IP, switching chips, and network interface cards (NICs) to the upper-layer management software, ScaleFabric achieves 100% self-developed capabilities, filling a gap in the domestic high-speed data center network field.

The ScaleFabric 400 series surpasses NVIDIA NDR in all aspects: the NICs use PCIe 5.0 interfaces, with port bandwidth up to 400Gbps and end-to-end communication latency as low as 0.9 microseconds; the switches have a single-port bandwidth of up to 800Gbps and a switching latency of only 260 nanoseconds. In terms of scalability, the interconnection scale of a single subnet is 2.33 times that of traditional InfiniBand, and the maximum cluster deployment scale can reach 114,000 NICs, thereby reducing the total network cost by 30%. Currently, scaleFabric has completed large-scale verification at the Zhengzhou core node of the National Supercomputing Internet, supporting the online operation of three 10,000-card clusters (totaling 30,000 cards), and has achieved stable operation for over 10 months at a scale of nearly 10,000 cards. Li Bin, Senior Vice President of Sugon, stated that the application of this product signifies that China's independent RDMA technology is gradually maturing. Wu Hequan, Academician of the Chinese Academy of Engineering, commented: "scaleFabric is the first domestically developed fully stacked 400G native RDMA high-speed network system. Its performance is comparable to mainstream international technologies, and its large-scale practical application verification fills the gap in China's high-speed network."

The advent of this product provides crucial independent technical support for large-scale AI model training and the construction of ultra-large-scale intelligent computing centers in China, breaking the long-standing foreign technological monopoly in the high-speed network field.