DGX B300 or DGX GB300 · scalable units · Mission Control
NVIDIA's validated, turnkey AI supercomputer, built from DGX B300 or DGX GB300 systems in standard scalable units and run with Mission Control.
Some organisations want to make as few design decisions as possible. DGX SuperPOD is for them. It is NVIDIA's own reference architecture for a complete AI supercomputer, specifying the DGX systems, the compute, storage and management networks, certified storage partners and the software, all validated together by NVIDIA.
The Blackwell Ultra generation comes in two flavours. DGX GB300 is the liquid-cooled rack-scale option with 72 GPUs per rack. DGX B300 is an air-cooled 10U system with eight GPUs, 2.1 TB of GPU memory and around 14 kW of power, for halls without liquid cooling. Both are deployed in scalable units, and published reference designs extend to well over 100 racks.
Our role is to make that blueprint real in your building: site readiness, logistics, integration with your identity and storage, and the operational processes that keep a SuperPOD productive for years.
Compute, fabric, storage and software are a tested combination. That removes most of the integration risk that slows down large clusters.
NVIDIA Mission Control handles job scheduling, health monitoring, failure detection and recovery across the whole SuperPOD, so a failed node costs minutes, not a weekend.
DGX B300 suits air-cooled halls; DGX GB300 brings rack-scale NVLink where liquid cooling is available. Both follow the same operational model.
Capacity is added a scalable unit at a time, with the fabric and storage sized for the next step from day one.
Building blocks of a Blackwell Ultra DGX SuperPOD (NVIDIA figures).
| DGX B300 system | 8 × Blackwell Ultra SXM, 2.1 TB GPU memory, Intel Xeon 6776P, 10U, ~14 kW |
|---|---|
| DGX B300 performance | FP4 144 | 108 PFLOPS, FP8 72 PFLOPS (sparse) |
| DGX B300 networking | 8 × ConnectX-8 (up to 800 Gb/s each), 2 × BlueField-3 DPUs (up to 400 Gb/s) |
| DGX GB300 rack | 72 Blackwell Ultra GPUs and 36 Grace CPUs, liquid cooled |
| Scalable unit (GB300) | 8 DGX GB300 racks |
| Software | DGX OS, NVIDIA AI Enterprise, NVIDIA Mission Control |
| Storage | NVIDIA-certified high-performance storage partners |
| Support | NVIDIA enterprise support, with Workstation as your UK point of contact |
Figures are NVIDIA's published specifications for the platform. Where NVIDIA quotes two Tensor Core figures, the first assumes sparsity and the second is dense. The exact build (CPUs, memory, storage, cooling, networking) is agreed with you and confirmed on a written quote.
A SuperPOD is planned with your facility team from the first conversation.
| Power | Around 14 kW per DGX B300; well over 100 kW per DGX GB300 rack |
|---|---|
| Cooling | Air with containment for DGX B300; direct liquid cooling for DGX GB300 |
| Space | Compute, network and storage racks for the full scalable unit, with room to add the next |
| Timeline | Supply, facility work and installation planned as one programme |
One accountable UK team from sizing to day-two operations
Supply, facility upgrades, installation and acceptance planned as one schedule with clear owners.
Power, cooling, cabling pathways and storage integration prepared before delivery.
Identity, storage, monitoring and network integration with your existing environment.
Mission Control operations, UK first-line support and escalation to NVIDIA.
NVIDIA data-centre GPUs are allocated by NVIDIA and the server makers, and lead times are measured in months rather than weeks. Complete, tested systems usually get allocation sooner than loose parts, so we quote these platforms as built systems. Shipments outside the UK, and many advanced GPUs within it, need an end-user and end-use check under UK and US export rules before we can confirm supply.
Trade members get our engineers' time on sourcing and design questions, early stock and waitlist priority, and free technical support on every order.
BasePOD is NVIDIA's reference architecture for smaller DGX deployments of a few systems. SuperPOD is the full-scale design for hundreds to thousands of GPUs, with stricter specifications for networking, storage and management.
SuperPOD uses NVIDIA-certified storage from partner vendors. We help you choose one that fits your existing estate and skills.
Tell us what you want it to produce, how many people will use it and where it would live. An engineer will come back with a sized recommendation and an honest view on whether owning makes sense for you.