NVIDIA HGX B200 8-GPU Baseboard (920-23686-2530-000)
Brand: NVIDIA | Category: GPUs
SKU: NVIDIA-920-23686-2530-000 | Part #: 920-23686-2530-000 | MPN: 920-23686-2530-000
Contact for Pricing — Request a Quote
About the NVIDIA HGX B200 8-GPU Baseboard (920-23686-2530-000)
The NVIDIA HGX B200 8-GPU Baseboard (920-23686-2530-000) delivers up to 20 petaFLOPS FP4 and up to 9 petaFLOPS FP8 tensor performance per GPU, with support for mixed-precision workloads spanning FP4, FP6, FP8, FP16, BF16, TF32, and FP64. Built on NVIDIA's Blackwell architecture and manufactured at TSMC 4NP, this 8-GPU system aggregates 1.4 TB of HBM3e memory and 8 TB/s memory bandwidth per GPU, enabling massive parallel compute for large-scale AI and HPC applications. The 2nd Generation Transformer Engine accelerates natural language processing and deep learning inference at scale.
Eight NVIDIA B200 GPUs interconnect via NVLink 5th Generation technology, providing 1.8 TB/s bidirectional GPU-to-GPU bandwidth across the baseboard. Multi-node scaling is supported through NVLink Switch integration for extended HPC domains. The HGX baseboard form factor is designed for OEM server integration and pairs with NVIDIA Quantum-X800 InfiniBand or NVIDIA Spectrum-X800 Ethernet for production deployments. For AI infrastructure teams and HPC architects, this platform delivers the memory density, interconnect speed, and precision flexibility required for transformer model training, inference serving, and scientific computing at enterprise scale. The CUDA, cuDNN, TensorRT, and NVIDIA AI Enterprise software stack ensures rapid deployment across existing infrastructure.
Enterprise Deployment Scenarios
- Large language model training and fine-tuning with unified 1.4 TB aggregate HBM3e capacity across all 8 GPUs
- High-throughput AI inference serving using FP8 and mixed-precision optimizations for latency-sensitive workloads
- Scientific computing and physics simulations leveraging FP64 support and NVLink 5th Generation bandwidth
- Multi-tenant cloud and data center deployments with NVIDIA Spectrum-X800 Ethernet and Quantum-X800 InfiniBand integration
- Containerized AI workloads via NVIDIA AI Enterprise runtime with native CUDA and TensorRT optimization
Contact Omnixon Global to request a detailed quote and technical consultation for the NVIDIA HGX B200 8-GPU Baseboard (920-23686-2530-000).
Technical Specifications
| Brand | NVIDIA |
| Category | GPUs |
| SKU | NVIDIA-920-23686-2530-000 |
| Part Number | 920-23686-2530-000 |
| Condition | New |
| Form Factor | SXM5 |
| Interface | NVLink 5.0 |
| Manufacturer Part Number | 920-23686-2530-000 |
| Product Name | NVIDIA HGX B200 8-GPU Baseboard |
| GPU Architecture | NVIDIA Blackwell |
| GPU Model per Slot | NVIDIA B200 |
| Number of GPUs | 8 |
| GPU Memory per GPU | 192 GB HBM3e |
| Total Aggregate GPU Memory | 1.4 TB HBM3e |
| Memory Bandwidth per GPU | 8 TB/s |
| GPU Interconnect | NVLink 5th Generation |
| NVLink Bidirectional Bandwidth | 1.8 TB/s (GPU-to-GPU, aggregate across baseboard) |
| FP4 Tensor Core Performance (per GPU) | Up to 20 petaFLOPS |
| FP8 Tensor Core Performance (per GPU) | Up to 9 petaFLOPS |
| Supported Precisions | FP4, FP6, FP8, FP16, BF16, TF32, FP64 |
| Transformer Engine Generation | 2nd Generation |
| Form Factor | HGX baseboard (OEM server integration) |
| Process Node | TSMC 4NP |
| Networking Compatibility | NVIDIA Quantum-X800 InfiniBand, NVIDIA Spectrum-X800 Ethernet |
| Software Stack | CUDA, cuDNN, TensorRT, NVIDIA AI Enterprise |
| Multi-Node Scaling | Supported via NVLink Switch for multi-node NVLink domains |
| Target Deployment | Enterprise data center, AI training clusters, HPC environments |
Frequently Asked Questions about NVIDIA HGX B200 8-GPU Baseboard (920-23686-2530-000)
What does the NVIDIA HGX B200 8-GPU Baseboard do?
The NVIDIA HGX B200 8-GPU Baseboard accelerates AI/ML training, inference, scientific HPC, and virtualization (vGPU) workloads. Typical deployments include LLM training clusters, computer-vision pipelines, financial risk modeling, and rendering farms.
What are the headline specs of the NVIDIA HGX B200 8-GPU Baseboard?
Key specifications for the NVIDIA HGX B200 8-GPU Baseboard: form factor SXM5; interface NVLink 5.0; 3 years support; new condition; support 3 years standard (manufacturer); raid support H100; gpu support NVIDIA B200; gpu count 8x; ai optimized Yes. Manufacturer part number 920-23686-2530-000. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.
Is the NVIDIA HGX B200 8-GPU Baseboard compatible with my infrastructure?
The NVIDIA HGX B200 8-GPU Baseboard requires a PCIe Gen4 or Gen5 x16 slot, server power adequate for the card's TDP, and CUDA/ROCm driver support in your hypervisor or bare-metal OS. Sales engineering will confirm chassis fit (1U/2U/4U), PCIe lane count, and PSU headroom before quoting.