Blackwell · air cooled · 8 GPUs per server
Air-cooled servers with eight RTX PRO 6000 Blackwell Server Edition GPUs each: the most accessible AI factory for enterprise agents, visual computing and physical AI.
Not every AI factory needs HBM and liquid cooling. Many enterprises want to run private agents, document search and retrieval, coding assistants and image or video models for hundreds of staff, and perhaps render, simulate or build digital twins on the same kit. The RTX PRO 6000 Blackwell Server Edition was made for exactly that mix.
Each GPU has 96 GB of GDDR7 and 4 PFLOPS of FP4 AI performance, and it can be split into up to four fully isolated instances so several teams or services share one card safely. Eight of them fit in an air-cooled server, giving 768 GB of GPU memory per node, and a reference rack of four nodes draws around 33 to 37 kW.
That means an AI factory you can host in many existing server rooms and colocation suites, built on the same NVIDIA AI Enterprise software as the largest systems. We also build our own Workstation 4U 8-GPU and 2U 4-GPU RTX PRO servers in the UK.
Serve 70B-class models, embeddings and rerankers side by side, close to your documents and systems, with no data leaving your network.
Fourth-generation RT cores and full graphics support mean the same GPUs render, stream virtual workstations and run Omniverse when they are not serving models.
Multi-Instance GPU splits each card into up to four isolated slices, so smaller services and teams get guaranteed capacity without buying whole GPUs.
Standard air-cooled servers and racks, with far lower power per rack than HBM platforms, make this the easiest AI factory to host.
A typical RTX PRO AI factory node and rack.
| GPU | NVIDIA RTX PRO 6000 Blackwell Server Edition, 96 GB GDDR7 with ECC, 1,597 GB/s |
|---|---|
| Rendimiento de IA | 4 PFLOPS FP4 per GPU |
| GPU power | Up to 600 W per GPU (configurable) |
| Partitioning | Up to four Multi-Instance GPU slices per card |
| Per server | 8 GPUs, 768 GB GPU memory, dual Xeon or EPYC |
| Red | Spectrum-X Ethernet with ConnectX SuperNICs or BlueField-3 |
| Escala | Typically 4 to 8 nodes per rack; up to 32 nodes (256 GPUs) per scalable unit |
| Software | NVIDIA AI Enterprise, NIM, Omniverse, Run:ai, Kubernetes |
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.
The friendliest AI factory to host.
| Rack power | Around 33.3 to 36.6 kW for four 8-GPU nodes (published reference figure) |
|---|---|
| Refrigeración | Air cooling with good containment; liquid-cooled GPU variants exist for dense builds |
| Rack | Standard deep server racks |
| Red | Ethernet leaf and spine, often integrated with your existing data-centre network |
One accountable UK team from sizing to day-two operations
We map your agents, retrieval, graphics and simulation workloads to GPUs, slices and servers.
Workstation RTX PRO servers built, burned in and imaged in the UK, or partner servers if you prefer a brand.
Kubernetes, NIM, vector database, identity and observability configured for self-service.
WSL Proxy at the edge, WSL Vault for secrets and SysOps247 monitoring.
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.
With 768 GB across eight GPUs, one server can serve models in the 70B to 200B+ parameter range at FP8 or FP4, or many smaller models at once.
Yes. Our 2U server takes up to four RTX PRO 6000 Server Edition GPUs, and the 4U server can start with two and grow to eight.
No. The Server Edition is passively cooled for server airflow and is the right card for racks. Workstation and Max-Q editions are designed for desk-side towers.
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.