Buy a server with 2× NVIDIA RTX PRO 6000 Blackwell Server Edition in Europe

The entry server with two Server Edition cards

You can buy a server with two NVIDIA RTX PRO 6000 Blackwell Server Edition cards from us, built to order and delivered anywhere in the EU. It is the entry configuration among our AI servers: two 96 GB cards, two AMD EPYC processors, 384 GB of DDR5 ECC memory and two NVMe drives. Our team sizes the configuration to your workload and budget and replies within one business day.

Two cards suit a company that moves its first AI service into production: a private assistant for a few hundred staff, RAG on its own documents, or a coding assistant for a development team. The server holds gpt-oss-120b as two independent copies, Llama 3.3 70B in FP8 as one pair split over both cards, or one model per card next to an embedding model and a reranker in MIG instances. With the example values of our private ChatGPT sizing guide, two copies of gpt-oss-120b cover the peak of about 500 employees. Larger models, or many departments on one server, call for four cards; a service that must survive a host failure calls for a second server.

Parts list of the entry configuration

The table shows the reference configuration and what we change per project.

Our AI servers page; NVIDIA-Certified Systems configuration guide (30 September 2026) for sockets, cores, memory, NVMe, Redfish and TPM; AMD's EPYC 9135 page for cores, read on 24 October 2026.

The host follows NVIDIA's configuration guide for NVIDIA-Certified Systems: at least two sockets, six physical cores per GPU, system memory of at least twice the total GPU memory and one NVMe drive per socket. For two 96 GB cards that is 12 cores, 384 GB and two drives. Two EPYC 9135 have 16 cores and 200 W each and 12 memory channels per socket, so 24 modules of 16 GB fill every channel. Their 32 cores also cover four cards, so the processors need no change if the server grows.

The two NVMe drives hold the operating system and the model files, mirrored so that the server keeps running when one drive fails. A store for several checkpoints of 60 to 170 GB, a RAG index or local logs calls for more drives in the free bays of the chassis. For users, 25 GbE carries chat traffic; 100 GbE pays off when models load from shared storage or the server later joins a second one behind a load balancer.

Options we change per project

Most of the variants come down to whether the server stays at two cards.

• 2U for two cards. The smallest version. Dell's PowerEdge R7725, for example, takes two cards in 2U, at 450 W each according to Dell's specification sheet. A chassis like this stays at two.
• A chassis for four. HPE's ProLiant DL385 Gen11 takes four cards at 600 W in 2U with its Maximum Performance fan kit, and HPE's DL380a Gen12 uses the same five power supplies for two or four double-width GPUs. Starting there costs more on day one and avoids a second server later, as our guide to adding GPUs to a server later explains.
• Memory for growth. Four cards need 768 GB by NVIDIA's rule. Twelve modules of 32 GB give the same 384 GB with half the channels free for the later step, at half the memory bandwidth until then.
• Power mode. With two cards the 450 W cap is not needed for the power feed. It lowers heat and is the rated mode in some 2U servers, at a cost of 2 to 9 per cent of throughput that StorageReview measured on the RTX PRO 6000 Workstation Edition.
• One processor. Single-socket 2U servers for two cards exist, but NVIDIA's current guide asks for two sockets, so our entry configuration uses two.
• Max-Q instead of Server Edition. The RTX PRO 6000 Max-Q is the workstation card, a 300 W blower card that NVIDIA supports up to four per workstation. In a rack server it belongs only in a chassis whose maker lists it, and many server makers do not; for continuous operation in a rack we recommend the Server Edition. For a desk-side system, see our page to buy a workstation with two RTX PRO 6000 Max-Q.

Models and conversations on two cards

By our users-per-card rule, 0.9 × 95.6 GiB − 3 GiB per card, the two cards hold 166.1 GiB of weights and KV cache. The table gives four examples; our guide to what a 2× RTX PRO 6000 Server Edition server runs has every layout, published results and the arithmetic.

Our estimates by the users-per-card rule: concurrent conversations at full context length, FP8 KV cache for Llama 3.3 70B, 16-bit for the others.

MIG gives the cards a second use. One card serves the main model whole while the other is split into four 24 GB instances for an embedding model, a reranker, speech-to-text and a small task model, each isolated from the others. A model split over both cards runs on whole cards, because NVIDIA's MIG guide states that NCCL is not supported with MIG; our RTX PRO 6000 MIG profiles show what fits in each size.

The figures count requests in flight at the same moment, which a gateway log shows at the busiest hour. We build this server in every variant above; send us your models and peak requests, and our team returns a configuration with the conversations per layout.

Rack, feed and air for about 1.8 kW

By our estimate, two cards at 600 W and two 200 W processors need about 1.8 kW at the wall, and about 1.5 kW with the cards at 450 W; the 15 per cent we add for memory, drives, fans and conversion losses is an assumption that a measurement on the delivered system confirms or replaces. We plan a 230 V 16 A single-phase feed at 80 per cent, which is 2.94 kW, so the server fits one such feed per side. With redundant power supplies on an A and a B feed, each side carries the whole server.

• Outlets. The power supplies decide the cord. HPE, for example, allows only C19 or C19-C20 cords for the 2,400 and 3,200 W supplies its DL380a Gen12 needs with this card, so check that the rack PDU has the matching outlets.
• Cooling. About 1.8 kW of heat needs roughly 290 CFM of air at an 11 °C rise, by the 160 CFM per kW in our guide to GPU rack power and cooling. NVIDIA's product brief asks for 43 CFM through each card's heat sink at a 25 °C inlet and 125 CFM at 45 °C.
• Room. Under full load the server belongs in a rack in a server room or data centre; our guide to a GPU server in an office covers the cases in between.
• Growth. Four cards at 600 W would need about 3.2 kW with the same processors, more than one 16 A feed; at the 450 W cap about 2.5 kW.

We check the rack position before we quote. Tell us the feed type, the A/B requirement and the inlet temperature, and we choose the chassis and power supplies to match.

Operating system, driver and licences on delivery

On request the server arrives with the operating system, NVIDIA driver, CUDA and a container runtime installed, so your team can start a serving engine on the first day.

Our AI servers page; driver and CUDA minimums from NVIDIA's product brief SP-12355-001_v02 for the Server Edition.

As of October 2026, R580 is NVIDIA's long-term branch with support to June 2028, and NVIDIA's security bulletin of 30 September 2026 names 580.178.04 as the first fixed R580 driver; our article on NVIDIA driver branches and CUDA versions explains the choice. Serving on bare metal with open-source engines needs no NVIDIA licence. NVIDIA AI Enterprise, licensed per GPU, is needed for NIM and for compute vGPU profiles; on vSphere the card needs ESXi 8.0 Update 3g or later.

From enquiry to delivery in the EU

The steps are the same for one server or a pair of them for failover.

1. You describe the workload: models, users, data, rack and timeline. Sizing is free of charge.
2. We propose the configuration and check that cards, chassis, power supplies and cooling fit your rack and feed.
3. You receive a written quote within one business day, with price and warranty terms per component; quantities and lead time are confirmed in writing with it.
4. You order on one EU contract with an EU invoice from Eurokommerz in Vienna; NVIDIA licences are lines on the same invoice.
5. We assemble the server and test it before delivery, with a power-on and basic load test.
6. We deliver anywhere in the EU, with commissioning on site on request; other destinations on request through our logistics partner.
7. Every component carries the manufacturer warranty, handled through us; DOA units are replaced, and the assembly is covered by us.

We also build in your own chassis or with parts you already own, after a compatibility check of the platform, power and cooling. For your own acceptance test on arrival, our GPU server acceptance testing guide lists the checks, and our GPU server quote comparison shows what each line of a quote should state.

Details that make the first quote exact

A few sentences are enough to start. These points let us quote the exact configuration in the first reply.

• The models you plan to run, with quantisation and the context length you will declare
• Peak requests in flight, or the number of users and how they work
• Whether the server should take four cards later, and when
• Other workloads on the same server: embedding and reranking, speech-to-text, virtual machines or MIG
• Software platform: bare metal, Kubernetes or vSphere, with the version
• Rack position: free height units, feed type and current, A/B feeds, PDU outlets and inlet temperature
• Delivery site, invoicing entity, timeline and any tender requirements, which our GPU server tender specification helps to word

Two servers are often the better first order when the assistant must stay up during maintenance, as our guide to LLM high availability on two GPU nodes explains.

Cards, servers and licences from one supplier

We supply the server with two NVIDIA RTX PRO 6000 Blackwell Server Edition cards in 2U or in a chassis for four, together with four-card and eight-card versions, and the RTX PRO 6000 cards on their own for servers you already run. NVIDIA AI Enterprise and vGPU licences come on the same EU contract and invoice as the hardware. The four-card version is on our page to buy a 4× RTX PRO 6000 Server Edition server.

FAQ