G6228 G3

The G6228 G3 is a powerful, ultra-scalable AI server built on Liantai’s next-generation AI hyper-converged architecture platform, designed for hyperscale data centers. It boasts the highest computing density, equipped with an NVIDIA Hopper-based HGX-8 GPU module in a 6U footprint. The system supports 4.0Tbps of network bandwidth, meeting the demands of parallel training of extremely large models with trillions of parameters. Equipped with two 4th-generation Intel® Xeon® Scalable processors (TDP 350W), it supports up to 4TB of system memory, 128TB of NVMe high-speed storage, and up to 12 PCIe Gen5 x16 expansion slots. It flexibly supports OCP 3.0, CX7, and various Smart NICs, creating a powerful AI computing platform for scenarios such as ultra-large model training, the metaverse, natural language understanding, inference, and AIGC.

Features

• Leading Architecture
  8 NVIDIA’s latest Hopper architecture GPUs;

  2 Intel® 4th Generation Xeon® Scalable processors;

  Integrating the Transformer engine greatly accelerates the training speed of large GPT models.

• Powerful performance
  Full PCIe 5.0 high-speed links within the node, quadrupling the CPU-to-GPU bandwidth;

  High-speed interconnection and expansion between nodes, 4.0Tbps non-blocking bandwidth IB/RoCE networking;

  Cluster-level optimized architecture design, GPU: Compute IB: Storage IB = 8:8:2.

• Ultimate energy efficiency
  Extremely optimized heat dissipation performance, and air duct decoupling design improves system energy efficiency by 20%;

  12V and 54V N+N redundant power supply separation design reduces power conversion loss.