Buy a server with 4× NVIDIA H200 NVL in Europe: 564 GB in one NVLink domain

Who buys a four-card H200 NVL server

You can buy a server with four NVIDIA H200 NVL cards from us, built to order and delivered anywhere in the EU. It is the four-card size among our AI servers: four cards of 141 GB HBM3e on NVLink bridges, two AMD EPYC processors, about 1.1 TB of DDR5 ECC memory and four NVMe drives. Our team sizes the configuration to your workload and budget and replies within one business day.

Four H200 NVL suit an organisation that needs more memory per model than PCIe cards give. A bank or insurer running a 70B model for several hundred concurrent requests, a research institute serving a 400B-class model in FP8, or a team fine-tuning a 70B model with LoRA all fit this size. The 4-way bridge joins the four cards into one NVLink domain of 564 GB, the largest an H200 NVL bridge allows. For a pilot or one model of up to 235B parameters, two cards with a 2-way bridge are enough, as our page to buy a 2× H200 NVL server describes. Models above about 490 GiB of weights, such as DeepSeek-V3.2 in FP8, need our training node with eight H200 NVL.

Configuration and options

The table shows the four-card configuration we start from and what we change per project.

Our AI servers page; NVIDIA AI Enterprise reference architecture (cores, memory floor, NVMe, network) and NVIDIA-Certified configuration guide (30 September 2026); HPE DL380a Gen12 QuickSpecs and Cisco UCS C845A M8 spec sheet A.11 for power supplies and memory, read on 24 October 2026.

We size the memory by NVIDIA's configuration guide for NVIDIA-Certified Systems: at least two sockets, six physical cores per GPU and system memory of at least twice the GPU memory, so 1,128 GB for four cards. Two EPYC 9005 processors have 24 memory channels, and 24 modules of 48 GB give 1,152 GB with one module per channel. NVIDIA's AI Enterprise reference architecture sets lower floors and stricter core counts: 128 GB per GPU, seven cores per GPU plus two per GPU for the operating system and two per MIG instance, which comes to 36 cores, or 92 with all 28 MIG instances. Our guide to CPU, RAM and storage for a 4-GPU server explains each tier.

Variants: bridge set, memory and room for eight

Three choices change the configuration.

• Bridge set. One 4-way bridge lets one copy of a model use all 561.6 GiB the driver reports, with TP=4 over NVLink. Two 2-way bridges give two pairs, each under one processor, which NVIDIA's product brief recommends; they suit two models or two copies. Cisco lists both sets for four cards in its UCS C845A M8, HPE lists the 2-way and 4-way bridges for the ProLiant Compute DL380a Gen12, and Dell's PowerEdge XE7740 guide uses one 4-way bridge for four cards. Our article on the H200 NVL NVLink bridge, 2-way or 4-way compares them.
• Memory. 1,152 GB covers serving. Cisco recommends 2,048 GB for four H200 NVL, at least three times the HBM, which pays off for fine-tuning with CPU offload or a KV cache spilled into system memory.
• Room for eight. A 4U chassis built for eight cards lets you add four cards and a second bridge set later. HPE requires five 2,400 or 3,200 W supplies for two or four double-wide GPUs and eight for eight, and Dell marks 512 GB and 1 TB of memory as not validated for eight H200 NVL in the XE7740, so the supplies and memory slots are planned at the start, as our guide to planning a GPU server for growth shows.

There is also a 2U option: Supermicro's single-socket SYS-212GB-NR, which takes up to four double-width cards including the H200 NVL. Its option list shows a GPU bay for two cards with NVLink, and NVIDIA's certified-systems guide asks for two sockets, so we recommend 4U for four H200 NVL. Our comparison of 2U and 4U for four cards covers the general case.

What it runs

By our users-per-card rule, 0.9 × 140.4 GiB − 3 GiB per card, the four cards hold 493.4 GiB of weights and KV cache. The table gives five examples; our guide to what a 4× H200 NVL server runs has the full model list, every layout 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 and Qwen3-235B, 16-bit for gpt-oss-120b.

These are requests in flight at the same moment, which a gateway log shows at the busy hour; the staff who use the service are many more. The H200 NVL has no FP4 arithmetic, so models are sized in FP8 or INT4. We supply the server in each layout above; send us your models and the peak requests in flight, and our team returns a configuration with the conversations per layout.

Power, cooling and the rack position

NVIDIA's product brief gives each card 600 W by default, 200 W minimum, with one 16-pin connector. By our estimate, four cards at 600 W with two AMD EPYC 9355 processors at 280 W need about 3.4 kW at the wall, and about 2.7 kW with the cards capped at 450 W. We plan a 230 V 16 A single-phase feed at 80 per cent of its rating, 2.94 kW, so the full-power server needs a 32 A single-phase position (5.89 kW planned) or a 400 V three-phase feed. With redundant supplies on an A and a B feed, each side carries the whole server.

• Outlets. HPE allows only C19 or C19-C20 cords for the 2,400 and 3,200 W supplies its DL380a Gen12 needs with the H200 NVL, so check that the rack PDU has C19 outlets.
• Cooling. The card is passive and takes its air from the server's fans; NVIDIA rates it for 10 to 45 °C ambient. About 3.4 kW of heat needs roughly 540 CFM at an 11 °C rise, by our guide to GPU rack power and cooling. Server makers set their own inlet limits; Dell gives 30 °C for eight H200 NVL in the XE7740.
• Room. A four-card server at full load belongs in a server room or data centre, as our guide to a GPU server in an office explains.

We check rack, feed and airflow before we quote; tell us the feed type, the A/B requirement and the inlet temperature at the rack position, and we choose the chassis and power mode to match.

Software and the included NVIDIA AI Enterprise

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 minimum "R565 TRD1 or later" from NVIDIA's product brief PB-12128-001_v01; MIG from NVIDIA's MIG user guide (13 October 2026).

NVIDIA's licensing guide, updated 2 September 2026, states that "Each NVIDIA H200 NVL Tensor Core GPU includes a five-year NVIDIA AI Enterprise subscription". The term starts 90 days after the board ships to the server maker, cannot be changed, and is activated with the GPU serial number on the system where the cards are installed; our article on activating NVIDIA AI Enterprise shows the steps. It covers NIM and compute vGPU (C-series); the H200 NVL has no graphics vGPU. Open-source engines such as vLLM run without it. As of October 2026, R580 is NVIDIA's long-term driver branch with support to June 2028, and NVIDIA's security bulletin of 30 September 2026 names 580.178.04 and 595.91.07 as fixed drivers. Models, RAG and the platform on top are a separate service, Private AI/ML.

Ordering, testing and warranty

The order runs in the same steps for one server or several.

1. You describe the workload: models, users, data, rack and timeline. Sizing is free of charge.
2. We propose the configuration, with the bridge layout, 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, lead time and warranty terms per component. Quantities and lead time are confirmed in writing with the quote.
4. You order on one EU contract with an EU invoice from Eurokommerz in Vienna.
5. The server is assembled and tested 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 supply the cards and bridges alone for a server you already run, after a check of its platform, power and cooling; our checklist for an H200 NVL in an existing server lists the points. To run your own acceptance test on delivery, use our GPU server acceptance testing guide, and to compare quotes line by line, our guide to GPU server quote comparison.

Details for the quote

A short description is enough to start; these points let us quote the exact configuration in the first reply.

• Models you plan to run, with precision (FP8 or INT4) and the context length you will declare
• Peak requests in flight, or the number of users and how they work
• Fine-tuning plans: method, model size and dataset
• One model across four cards, or several models or copies, which decides the bridge set
• 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
• Whether the server should grow to eight cards, delivery site, invoicing entity and timeline; for a tender, our GPU server tender specification helps with the wording

Send these points with the form below, and our team replies with a configuration and a written quote within one business day.

What we supply

We supply the server with four NVIDIA H200 NVL and their NVLink bridges in 4U, with one 4-way bridge or two 2-way bridges, as well as two-card and eight-card versions and the H200 NVL cards and bridges for servers you already run. The five-year NVIDIA AI Enterprise subscription comes with each card, and further NVIDIA AI Enterprise licences come on the same EU contract and invoice as the hardware.

FAQ