Home high speed internet

Building a network with cores to accelerate smart computing in China

When large models enter the Wanka cluster era, network performance has become a key bottleneck that determines computing efficiency. Sugon ScaleFabric high-speed network is specially designed for large-scale AI clusters. Based on fully self-developed chips and native lossless RDMA network technology, ScaleFabric provides the ultimate throughput with zero packet loss and ultra-low latency, allowing every bit of computing power to be fully accelerated, building an autonomous and controllable high-speed base for intelligent computing centers.

Our Advantages

  • Full-stack In-house

    China's first InfiniBand-class high-speed network system, with 100% in-house design from 112G SerDes IP and network chips through management software.

    Learn More >
  • Extreme Performance

    800G per link, end-to-end NIC latency under 1 ?s, and switch latency around 260 ns.

  • Extreme Scale

    Compared with NDR, a single subnet can scale to 114,623 cards, supporting scaleX 10k-card clusters, national supercomputing internet hubs, and 60k-card AI4S clusters.

  • Ecosystem Compatible

    Native Verbs interfaces and broad communication-library support, compatible with HPC and AI applications.

Industry Focus

  • News

    Sugon scaleFabric: China-developed native RDMA high-speed network launch

    Sugon March 2026

    Learn More >
  • News

    Xinhuanet | Winning the AI compute scale battle, advancing high-speed interconnect for China-developed clusters

    Sugon March 2026

    Learn More >
  • News

    Historic breakthrough: Sugon 10,000-GPU supercluster sets a 41.47 billion-atom DFT simulation record

    Sugon March 2026

    Learn More >

Categories

Success Stories

News Center

Downloads

Contact Us

After-Sales Service

Solemn Statement