The Ultimate AI Training Appliance by NVIDIA
The sixth-generation DGX datacentre AI appliance is built around the Blackwell architecture and the flagship B200 accelerator, providing unprecedented training and inferencing performance in a single system.
The DGX B200 includes 400Gb/s Connect-X7 Smart NICs and Bluefield DPUs for connecting to external storage, supported by the NVIDIA Base Command management suite and the NVIDIA AI Enterprise software stack, backed by specialist technical advice from NVIDIA DGXperts.

Accelerated AI with NVIDIA B200 GPUs
The latest DGX B200 appliance features eight SXM6 B200 Tensor Core GPUs, featuring a total of 1.44TB of memory, delivering up to 15x the performance of the DGX H100. The B200 GPUs are connected by 5th gen NVLink technology, creating the fastest AI platform available today.

NVIDIA DGX B200 | ||
---|---|---|
GPUs | 8x B200 | |
FP4 TENSOR CORE** | 144 PFLOPS | 72 PFLOPS | |
FP8/FP6 TENSOR CORE* | 72 PFLOPS | |
INT8 TENSOR CORE* | 72 POPS | |
FP16/BF16 TENSOR CORE* | 36 PFLOPS | |
TF32 TENSOR CORE* | 18 PFLOPS | |
FP32 | 600 TFLOPS | |
FP64/FP64 TENSOR CORE* | 296 TFLOPS | |
ARCHITECTURE | Blackwell | |
CUDA CORES | TBC | |
TENSOR CORES | TBC | |
TOTAL MEMORY GPU | 1.4TB HBM3e | |
MEMORY CONTROLLER | 4,096-bit | |
NVLINK | 5th gen | |
NVSWITCH | NVLink 5 Switch | |
NVSWITCH GPU-TO-GPU BANDWIDTH | 1.8 TB/s | |
NVLINK BANDWIDTH | 14.4 TB/s |
* With sparsity
** With sparsity | without sparsity
AI-Ready Software Stack
NVIDIA Base Command powers the DGX systems, enabling organisations to leverage the best of NVIDIA software innovation. Enterprises can unleash the full potential of their DGX infrastructure with a proven platform that includes enterprise-grade orchestration and cluster management, libraries that accelerate compute, storage and network infrastructure, and an operating system optimised for AI workloads. This is further enhanced by NVIDIA AI Enterprise.

NVIDIA AI Enterprise
NVIDIA AI Enterprise unlocks access to a wide range of frameworks that accelerate the development and deployment of AI projects. Leveraging pre-configured frameworks removes many of the manual tasks and complexity associated with software development, enabling you to deploy your AI models faster as each framework is tried, tested and optimised for NVIDIA GPUs. The less time spent developing, the greater the ROI on your AI hardware and data science investments.
Rather than trying to assemble thousands of co-dependent libraries and APIs from different authors when building your own AI applications, NVIDIA AI Enterprise removes this pain point by providing the full AI software stack including applications such as healthcare, computer vision, speech and generative AI.
Enterprise-grade support is provided, 9x5 with a 4-hour SLA with direct access to NVIDIA’s AI experts, minimising risk and downtime, while maximising system efficiency and productivity. A three-year NVIDIA AI Enterprise license is included with 3XS AI workstations with A800 GPUs as standard. You can also purchase one, three- and five-year licenses with other GPUs.
Workload Management
Run:ai software allows intelligent resource management and consumption so that users can easily access GPU fractions, multiple GPUs or clusters of servers for workloads of every size and stage of the AI lifecycle. This ensures that all available compute can be utilised and GPUs never have to sit idle. Run:ai’s scheduler is a simple plug-in for Kubernetes clusters and adds high-performance orchestration to your containerised AI workloads.
FIND OUT MORE
AI Optimised Storage
AI Optimised storage appliances ensure that your NVIDIA DGX systems are being utilised as much as possible and always working at maximum efficiency. Scan AI offers software-defined storage appliances powered by PEAK:AIO and further options from leading brands such as Dell-EMC, NetApp and DDN to ensure we have an AI optimised storage solution that is right for you.
FIND OUT MORE
Managed Hosting Solutions
AI projects scale rapidly and can consume huge amounts of GPU-accelerated resource alongside significant storage and networking overheads. To address these challenges, the Scan AI Ecosystem includes managed hosting options. We’ve partnered with a number of secure datacentre partners to deliver tailor-made hardware hosting environments delivering high performance and unlimited scalability, while providing security and peace of mind. Organisations maintain control over their own systems but without the day-to-day admin or complex racking, power and cooling concerns, associated with on-premise infrastructure.
FIND OUT MORE
NVIDIA BasePOD & SuperPOD
DGX BasePOD and SuperPOD are NVIDIA reference architectures based around a specific infrastructure kit list that Scan can configure and deploy into your organisation to deliver AI at scale.
PODs start as small as 20 DGX nodes, scaling all the way to 140 nodes, managed by a comprehensive software stack to form a complete cluster.

Start your DGX Journey
The NVIDIA DGX B200 AI is available with three- and five-year support contract, extendable at a later period. There are also comprehensive media retention packages available for more data-sensitive projects.
GPU: 8x NVIDIA B200 80GB – 1.44TB total
CPU: 2x Intel Xeon Platinum 8570 – 112 cores / 224 threads total
RAM: 4TB ECC Reg DDR5
System Drives: 2x 1.92TB NVMe SSDs
Storage Drives: 8x 3.84TB NVMe SSDs
Networking: 8x 400Gb/s NVIDIA ConnectX-7 InfiniBand/Ethernet and 2x 400Gb/s NVIDIA Bluefield DPUs InfiniBand/Ethernet
Power: 14.3kW
Form Factor: 10U

Frequently Asked Questions (FAQ)
The NVIDIA DGX B200 is a powerful AI supercomputer designed for enterprise and research use. It features 8 NVIDIA Blackwell GPUs, up to 4 TB of system memory, and high-speed networking, making it ideal for training and deploying large language models, generative AI, advanced inference, and scientific computing. It includes the full NVIDIA AI software stack (additional charge may apply) and is built for on-premise AI infrastructure as part of DGX BasePOD and SuperPOD solutions.
The DGX B200 is a next-generation upgrade over the DGX H100 & H200, offering major improvements in performance, memory, and AI model handling:
Feature | H100 (Hopper) | H200 (Hopper Refresh) | B200 (Blackwell) |
---|---|---|---|
GPU Architecture | Hopper | Hopper (with HBM3e) | Blackwell |
Launch Year | 2022 | 2024 | 2024 (announced), shipping 2H 2024 |
FP8 Training Perf. | ~32 petaFLOPS (DGX H100) | Same as H100 | ~72 petaFLOPS (DGX B200) |
FP4 Inference Perf. | N/A | N/A | ~144 petaFLOPS (DGX B200) |
Memory Capacity | 80 GB HBM3 | 141 GB HBM3e | 192 GB HBM3e |
Memory Bandwidth | 3.35 TB/s | 4.8 TB/s | ~6 TB/s |
Transformer Engine | 1st-gen | 1st-gen | 2nd-gen with FP4/FP8 support |
NVLink Support | Yes (NVLink 4) | Yes | Yes (faster, 1.8 TB/s via NVSwitch) |
Target Use Case | AI training, HPC | Bigger models, more memory-bound | Massive LLMs, GenAI, real-time inference |
Summary:
- H100: Great for general AI training and HPC, computer vision NLP and deep learning. Offers an excellent balance between compute power and memory for most AI workloads.
- H200: Boosted memory version of H100; ideal for larger, memory-bound models like LLMs, memory-intensive AI inference and foundational model tuning.
- B200: The next-gen flagship; ideal for training and deploying large-scale LLMs, generative AI at scale and multi-GPU systems.
The DGX B200 is optimised for demanding AI workloads such as large-scale deep learning training, generative AI (including LLMs and diffusion models), high-performance inference, data analytics, and scientific computing. With its 8 Blackwell GPUs, ultra-fast memory, and high-speed networking, it's especially well-suited for enterprises and research institutions developing foundation models, deploying real-time AI applications, or running complex simulations.
Yes. Scan is an official UK NVIDIA Elite Partner offering the DGX B200 with full configuration, deployment, and support services. Scan is a DGX MSP (Managed Services Provider) that enables businesses to access the power of DGX infrastructure, like the DGX B200, without needing the skillset on-site to oversee the infrastructure. Instead, DGX MSPs provide full wraparound services to assist with anything from the initial deployment & configuration through to on-going services that encompass support, updates and overall platform management.
The NVIDIA DGX B200 is a high-performance AI system featuring 8× NVIDIA Blackwell B200 GPUs with 1,440 GB of HBM3e memory, delivering up to 72 petaFLOPS of AI training performance. It includes dual Intel Xeon Platinum CPUs, up to 2 TB of system RAM (expandable to 4 TB), and a mix of NVMe SSD storage for OS and data caching. The system supports high-speed networking with up to 400 Gb/s InfiniBand or Ethernet and comes with advanced software tools including NVIDIA AI Enterprise and Base Command. It fits into a 10U rack space and consumes up to 14.3 kW of power.
Scan provides full deployment services, including consultation, infrastructure assessment, rack integration, and ongoing support, tailored for enterprise and research environments.
Yes, the DGX B200 is purpose-built for on-premise AI infrastructure. It delivers powerful performance for training, inference, and analytics with 8 Blackwell GPUs, high-bandwidth NVSwitch interconnects, and up to 4 TB of system memory. Its 10U rackmount form factor, enterprise-grade networking (up to 400 Gb/s), and support for NVIDIA AI Enterprise software make it ideal for datacenters and organisations deploying advanced AI workloads on-site. However, if you can’t host this internally, we have several data centre partnerships in place to assist with co-location services as part of a complete services contract with SCAN.
The DGX B200 is ideal for CTOs, AI infrastructure leads, data centre architects, and researchers in enterprise, healthcare, academia, and government sectors needing cutting-edge performance for training and deploying AI models.
The Blackwell GPU architecture is designed specifically for next-generation AI workloads, offering major advances in performance, efficiency, and scalability. Key features include:
- Transformational AI performance: Supports FP4 and FP8 formats for massive throughput—ideal for training and inference of large language models and generative AI.
- Second-gen Transformer Engine: Accelerates training and inference of transformer-based models more efficiently than previous architectures.
- Enhanced NVLink bandwidth: Enables faster GPU-to-GPU communication for multi-GPU systems like the DGX B200.
- Advanced memory architecture: Uses high-capacity HBM3e memory for up to 1.5 TB/s bandwidth per GPU.
- Security and reliability: Includes confidential computing and RAS (reliability, availability, and serviceability) features to support enterprise and mission-critical environments.
These innovations make Blackwell GPUs particularly well-suited for ultra-large-scale AI models, delivering breakthrough performance in both compute and energy efficiency.
Yes. Scan can assess your current infrastructure and deliver a fully compatible DGX B200 deployment plan, including cooling, networking, and power considerations.
Absolutely. The DGX B200 is built for scalable AI infrastructure and can be clustered into DGX BasePODs or DGX SuperPODs. By interconnecting multiple DGX B200 systems using NVIDIA Quantum-2 InfiniBand or Ethernet with ConnectX-7 and BlueField-3, organisations can build AI datacenters capable of training and deploying the largest generative AI models and LLMs. These clusters benefit from the DGX B200’s ultra-fast NVLink and NVSwitch architecture for multi-node GPU communication, ensuring minimal bottlenecks at scale.
The DGX B200 is purpose-built to accelerate generative AI and LLM workloads at scale. It features 8 Blackwell GPUs with a combined 1,440 GB of HBM3e memory and up to 72 petaFLOPS of AI training performance, enabling it to train and fine-tune massive models efficiently. The system's second-gen Transformer Engine, high-bandwidth NVSwitch interconnects, and support for FP4/FP8 precision allow for faster training and inference of transformer-based architectures. Integrated with the NVIDIA AI software stack, the DGX B200 offers a complete platform for developing, deploying, and managing LLMs and generative AI applications both on-premise and in hybrid environments..
The NVIDIA DGX B200 is equipped with advanced networking features designed to meet the demands of high-performance AI workloads:
- High-Speed Connectivity: It includes 4 OSFP ports accommodating 8 single-port NVIDIA ConnectX-7 VPI adapters, each supporting up to 400 Gb/s for both InfiniBand and Ethernet connections.
- Data Processing Units (DPUs): The system features 2 dual-port QSFP112 NVIDIA BlueField-3 DPUs, each capable of up to 400 Gb/s, facilitating efficient data handling and network offloading.
- Management Networking: : For system management, it provides a 10 Gb/s onboard NIC with RJ45, a 100 Gb/s dual-port Ethernet NIC, and a host baseboard management controller (BMC) with RJ45.
These networking capabilities ensure that the DGX B200 can handle intensive data transfer requirements, making it suitable for scalable AI deployments and integration into high-performance computing environments.
As pricing depends on configuration and support options, contact Scan’s enterprise team for a tailored quote and expert consultation.