NVIDIA DGX B200 8x B200 SXM6 Server System

The NVIDIA DGX B200 is a purpose-built AI supercomputer integrating eight NVIDIA B200 SXM6 GPUs into a dense, rack-optimized 10U chassis. Built on the NVIDIA Blackwell GPU architecture, each B200 GPU delivers up to 20 petaFLOPS of FP4 AI compute, interconnected via NVLink 5 at 1.8 TB/s aggregate GPU-to-GPU bandwidth. The system is engineered to handle the most demanding large-scale AI training and inference workloads, offering a unified, software-optimized platform that ships with the NVIDIA AI Enterprise software stack, CUDA, and cuDNN libraries pre-validated and ready for deployment.

At the heart of the DGX B200 is the Blackwell GPU architecture, which introduces second-generation Transformer Engine capabilities, a dedicated FP8 and FP4 compute path, and a fifth-generation NVLink fabric. The eight B200 SXM6 GPUs share a high-bandwidth NVSwitch fabric, enabling any-to-any GPU communication without bottlenecking through the host CPU. Each B200 GPU carries 192 GB of HBM3e memory, giving the full system 1.5 TB of aggregate GPU memory — sufficient to host frontier-scale large language models and multimodal foundation models entirely within GPU memory during both training and inference phases.

The DGX B200 is positioned as a full-stack enterprise AI infrastructure node, designed for deployment in modern air-cooled or direct liquid-cooled data center environments. It integrates dual Intel Xeon Scalable host CPUs, high-capacity system DRAM, and multiple high-speed network interface options to sustain multi-node cluster scaling. NVIDIA's DGX software stack, including Base OS, system management, and performance benchmarking tools, is validated and supported on the platform, enabling enterprise IT teams to achieve production-grade AI infrastructure with a controlled, auditable software bill of materials.

Ideal for

• Training and fine-tuning frontier-scale large language models (LLMs) with hundreds of billions of parameters entirely within the system's 1.5 TB aggregate GPU memory pool
• High-throughput generative AI inference serving for real-time enterprise applications including retrieval-augmented generation (RAG), code generation, and multimodal AI
• Accelerated drug discovery and computational biology workloads including molecular dynamics simulation and protein structure prediction at scale
• Enterprise computer vision and video analytics pipelines requiring sustained multi-stream inference throughput across GPU-accelerated preprocessing and model execution
• Multi-node distributed deep learning research clusters where multiple DGX B200 nodes are networked via InfiniBand or Ethernet to form a tightly coupled AI supercluster
• Sovereign AI and national AI infrastructure deployments requiring on-premises, air-gapped, or compliance-controlled environments with full-stack NVIDIA software support

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.