NVIDIADGX B200
Enterprise-grade unified AI platform: the solid foundation for building an AI factory
Powered by 8 NVIDIA Blackwell GPUs and connected via fifth-generation NVLink, it delivers 3x faster training and 15x faster inference than the previous-generation DGX H00, covering the entire AI workflow from development to deployment. Accelerate AI transformation for enterprises of all sizes and stages.
One platform accelerationEnterprise AI Transformation
The NVIDIA DGX B200 is a unified AI platform built on the NVIDIA Blackwell architecture, designed to support diverse workloads such as large language models, recommendation systems, and chatbots. It covers the entire AI workflow—from data preparation and training to fine-tuning and inference. As a foundational building block for DGX BasePOD and DGX SuperPOD, it enables enterprises to elastically scale their AI infrastructure on demand, running AI workloads at the scale of hyperscale cloud providers.
UnifiedAI Factory
From unified development and deployment to Blackwell performance leaps, from mature infrastructure standards to high-speed interconnects, DGX B200 delivers complete capabilities for enterprise AI deployment.
All-in-one platform from development to deployment
A single platform covering the full AI workflow—from data preparation and training to fine-tuning and inference. It provides enterprise development teams with a unified, accelerated compute foundation, streamlining the path from AI application development to deployment.
Blackwell Architecture: A Performance Leap
Powered by NVIDIA Blackwell architecture, it delivers 3x faster training and 15x faster inference compared to DGX H00, efficiently supporting diverse workloads such as large language models, recommendation systems, and chatbots.
Mature and reliable infrastructure standards
A fully optimized hardware platform that runs the complete NVIDIA AI software stack out of the box, including NVIDIA Mission Control and NVIDIA AI Enterprise, with rich third-party ecosystem support and professional services.
Fifth-Generation NVLink High-Speed Interconnect
8 Blackwell GPUs interconnected via fifth-generation NVLink and dual NVSwitches deliver 14.4 TB/s aggregate bandwidth. Integrated with ConnectX-7 and BlueField-3 networking, this architecture supports large-scale distributed training and inference.
NVIDIA DGX B200Technical Specifications
Complete parameter overview. Contact our technical consultant for selection advice.
| GPU | 8x NVIDIA Blackwell GPUs |
| GPU VRAM | 1,440 GB total capacity, 64 TB/s HBM3e bandwidth |
| Performance | FP4 Tensor Core: 144 PFLOPS | 72 PFLOPS* FP8 Tensor Core: 72 PFLOPS** |
| NVSwitch | 2x |
| NVLink Bandwidth | 14.4 TB/s aggregated bandwidth |
| CPU | 2x Intel Xeon Platinum 8570 Processors 112 cores, 2.1 GHz (base) / 4 GHz (max turbo) |
| System Memory | 2 TB, configurable up to 4 TB |
| Network | 4 OSFP ports corresponding to 8 single-port NVIDIA ConnectX-7 VPI Supports NVIDIA InfiniBand/Ethernet up to 400 Gb/s 2 dual-port QSFP112 NVIDIA BlueField-3 DPUs Supports NVIDIA InfiniBand/Ethernet up to 400 Gb/s |
| Manage Network | 10 Gb/s onboard network interface card (NIC), RJ45 port 100 Gb/s Dual-Port Ethernet NIC Host Baseboard Management Controller (BMC), RJ45 port |
| Storage | Operating System: 2x 1.9 TB NVMe M.2 Internal Storage: 8x 3.84 TB NVMe U.2 |
| System Power Consumption | Approx. 14.3 kW (max) |
| Software | NVIDIA DGX OS / Ubuntu (Included) NVIDIA AI Enterprise (Recommended) NVIDIA Mission Control (Recommended) |
| Rack Unit | 10U |
| System Size | Height: 17.5 inches (444 mm) Width: 19.0 inches (482.2 mm) Length: 35.3 inches (897.1 mm) |
| Operating Temperature | 10-35°C（50-90°F） |
| Support | 3-Year Business Standard Support Service |
* Specifications are displayed in Sparse | Dense modes; ** is in Sparse mode, with dense performance at half the rate of sparse specifications. Specs sourced from NVIDIA's official product page; actual configurations may vary by unit.
Product BrochureDownload
Below is a preview of the NVIDIA DGX B200 product brochure. Click the button to generate and download the PDF version for internal review and selection archiving.
DGX B200 features 8 NVIDIA Blackwell GPUs interconnected via 5th-generation NVLink and dual NVSwitch, delivering 144 PFLOPS FP4 (sparse) compute, 1,440GB total GPU memory, and 14.4 TB/s aggregate bandwidth—representing a 3x improvement in training performance and a 15x boost in inference over DGX H100. Its 1U rack-mounted chassis comes preloaded with the full NVIDIA AI software stack, serving as the foundational building block for DGX BasePOD and DGX SuperPOD.
| GPU | 8x NVIDIA Blackwell GPUs |
| GPU VRAM | 1,440 GB HBM3e (64 TB/s) |
| FP4 Performance | 144 PFLOPS (sparse) | 72 PFLOPS (dense) |
| FP8 Performance | 72 PFLOPS (sparse) |
| NVLink Bandwidth | 14.4 TB/s aggregated bandwidth (2x NVSwitch) |
| CPU | 2x Intel Xeon Platinum 8570 (112 cores) |
| System Memory | 2 TB (configurable up to 4 TB) |
| Network Computing | 8x ConnectX-7 VPI + 2x BlueField-3 DPU (up to 400 Gb/s) |
| Storage | 2x 1.9TB M.2 + 8x 3.84TB U.2 NVMe |
| Power / Rack Unit | Approx. 14.3 kW (max) / 1 0U |
Brochures are generated in real-time from page content. If your browser blocks the download, allow pop-ups and try again.
Ready to deploy DGX B200?
HuaYing Technology provides end-to-end professional services for DGX B200 selection consulting, cluster planning, and deployment implementation.
Purchase Intent
Submit your requirements and contact details. A sales consultant will confirm pricing, stock availability, and delivery options with you. Product prices are subject to quotation.