SuperMicro SYS-422GS-NB3RT-ALC

Example Use Cases:

• Large-scale LLM training
  The system is built around 8x NVIDIA HGX B300 GPUs, 5th-gen NVLink at 1.8 TB/s, and 2.3 TB of HBM3e GPU memory per system, which is exactly the kind of configuration needed for training large foundation models efficiently.

• Multi-modal AI model training
  Supermicro explicitly calls out “LLMs & Multi-modal LLMs,” and the combination of 8 GPUs plus up to 8 TB of DDR5 memory makes it well suited for training models that combine text, vision, audio, or video inputs.

• Deep learning inference at scale
  The system is also explicitly positioned for “Deep Learning Inferencing,” and its dense GPU footprint plus 8x ConnectX-8 SuperNICs up to 800 Gb/s make it a strong option for high-throughput inference clusters serving demanding models.

• General enterprise AI / accelerated AI infrastructure
  Supermicro lists “Artificial Intelligence” as a primary application, and the platform’s liquid-cooled design, redundant 6.6 kW power supplies, and hot-swap NVMe storage make it appropriate as a production-grade AI node in enterprise or hosted environments.

• Distributed AI cluster node for multi-node training
  Even beyond single-node workloads, this server is a strong fit as a scale-out training node because it combines 8 onboard GPUs with 8 OSFP 800 Gb/s InfiniBand ports and 4 PCIe Gen5 x16 slots, which is ideal for tightly coupled multi-node AI clusters.

  Download Data Sheet