Buy a 2U server with 4× NVIDIA RTX PRO 4500 Blackwell Server Edition in Europe

A four-card server for a department or a VDI pilot

You can buy a 2U server with four NVIDIA RTX PRO 4500 Blackwell Server Edition cards from us, built to order and delivered anywhere in the EU. It gives four separate 32 GB GPUs, eight MIG instances of 16 GB or up to 64 virtual desktops at 660 W of board power, so it runs on one ordinary 230 V 16 A feed. Our team sizes the configuration to your workload and budget and replies within one business day.

The server suits organisations that need GPUs for one department or one service. A design or engineering department moves its CAD users to virtual workstations on one host. An IT team runs a first vGPU pilot before it orders a cluster. A business unit serves an assistant model, an embedding model and a reranker for its document search. A university chair gives students and researchers isolated 16 GB GPUs instead of a shared card. Each of these fits in two height units on one standard feed.

Four cards are also the entry point for companies that expect to grow. Supermicro lists eight of these cards only in 3U, 4U and 5U chassis, so a company that wants to add cards later can start with four cards in an eight-card chassis and accept more rack space from the first day.

The server is the wrong choice for one 70B-class model with many users. Llama 3.3 70B in FP8 takes all four cards as one copy here, while one RTX PRO 6000 Server Edition holds it alone; our page to buy a 4× RTX PRO 6000 Server Edition server covers that build. Training and full fine-tuning belong on H200 NVL with NVLink bridges.

Parts of the four-card build

NVIDIA's configuration guide for NVIDIA-Certified Systems, updated on 30 September 2026, sets the floor for a GPU server: two sockets, at least six physical cores per GPU and system memory of at least twice the GPU memory. For four cards of 32 GB that is 24 cores and 256 GB.

NVIDIA-Certified Systems configuration guide (30 September 2026); Supermicro "RTX PRO Blackwell Server Edition" systems page and 2CRSi RTX PRO 4500 page for chassis and processors; our AI servers page; all read on 24 October 2026.

The chassis decides the processor family. On Supermicro's page, the 2U SYS-222H-TN and SYS-222C-TN take four cards with two Intel Xeon 6700 processors. 2CRSi, a French server maker, lists up to four cards in its 2U Mona 1.4SR with two AMD EPYC 9004. Lenovo's option guide for the card, updated on 28 July 2026, gives the power cabling for four cards in the 2U ThinkSystem SR665 V3. With two AMD EPYC 9355 of 32 cores and 12 memory channels each, 24 modules of 16 GB give 384 GB with one module per channel.

Supermicro "RTX PRO Blackwell Server Edition" systems page, 2CRSi RTX PRO 4500 Blackwell page and Lenovo Press LP2391 (updated 28 July 2026), all read on 24 October 2026.

Two of these listings show why the single-slot card matters. The SYS-222C-TN takes four RTX PRO 4500 Server Edition cards but only two RTX PRO 6000 Server Edition, and the SYS-222H-TN takes four of the smaller card and none of the larger one. The maker's configurator and GPU population rules decide risers, fans and cables for any model, and Lenovo states that the card "does not ship with auxiliary power cables" because the cables are server-specific, so they belong in the server order. Our guide on 2U or 4U for four cards explains what changes with the height.

Three ways to configure it

The hardware is nearly the same in all three; the software, the licences and the system memory change.

• Inference host for a department. Bare metal or Kubernetes with one serving engine per model. By our users-per-card rule, each card holds about 25.7 GiB of weights and cache, so a model up to about 20 GiB of weights runs as one copy per card, and larger ones as TP=2 pairs or one TP=4 copy.
• VDI host. vGPU 20 on VCF 9.0.1 or later, ESXi 8.0 Update 3g or later, RHEL KVM or Ubuntu, with vPC or RTX vWS licences. NVIDIA's RTX vWS sizing guide gives light users 4 vCPUs and 8 to 12 GB of RAM each and medium users 8 vCPUs and 16 to 24 GB, so 32 light users need up to 384 GB for their virtual machines; we size the processors and memory from your seat counts.
• Mixed host with MIG. Some cards split into two 16 GB halves for embedding models, rerankers or student environments, the others stay whole for a larger model. MIG is set per card, and a split model needs whole cards.

NVIDIA's product brief allows a power cap from 100 W to 165 W per card, as our RTX PRO 4500 Server Edition guide cites. At 100 W the four cards draw 400 W, which helps in a rack position with little power left. We supply the server in each of these variants; send us your models, seat counts or courses, and our team replies with a configuration and the capacity it gives.

Capacity at a glance

By our users-per-card rule, 0.9 × the memory the driver reports less 3 GiB per card, the four cards hold about 102.8 GiB of weights and cache. Our guide to 4× RTX PRO 4500 Server Edition in 2U has every layout, the derivation and the published single-card measurements.

Our estimates by the users-per-card rule, conversations at full length with an FP8 KV cache; desktops from NVIDIA's vPC sizing guide and users from NVIDIA's RTX vWS sizing guide (both updated 19 August 2026), multiplied by four.

The conversations are requests in flight at the same moment. At the example values of our guide to a private ChatGPT server by company size, about 20 requests are in flight for 500 staff and 40 for 1,000, so 64 conversations on a 14B model cover a department of 1,000 with room to spare. The desktop figures are NVIDIA's maximum densities; a pilot with your own desktops sets the profile.

Rack, feed and cooling for about 1.4 kW

By our estimate, four cards at 165 W and two AMD EPYC 9355 at 280 W each draw about 1.4 kW at the wall, including 15 per cent for memory, drives, fans and conversion losses. Processors with a higher TDP raise the figure; a measurement on the delivered system confirms it.

• Feed. A 230 V single-phase feed at 16 A is rated at 3.68 kW, and we plan 80 per cent of it, 2.94 kW. One server leaves more than half of that free. Two of them on one feed would draw about 2.8 kW by our estimate, too close to the limit to plan without measured figures. Four RTX PRO 6000 Server Edition cards at 600 W, by comparison, need about 3.4 kW and do not fit one such feed.
• Redundancy. With a redundant pair of power supplies on A and B feeds, each side carries the whole server. The supply rating comes from the chassis maker's rules for the GPU configuration.
• Cooling. About 1.4 kW of heat needs roughly 220 CFM of air at an 11 °C rise, by the 160 CFM per kW in our guide to GPU rack power and cooling. The card has no fan; Lenovo describes it as a passive card "requiring system airflow to operate", and the server maker sets the inlet limit for the chassis with this GPU.
• Room. A 2U GPU server under load is loud and belongs in a server room or data centre, as our guide to a GPU server in an office explains.

Software on delivery

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

Our AI servers page; driver branch as cited in our RTX PRO 4500 Server Edition guide; NVIDIA vGPU documentation, read on 24 October 2026.

NVIDIA's vGPU documentation lists vGPU 20 on the R595 branch with an end-of-life date of March 2027, while vGPU 19, the long-term branch supported until July 2028, does not cover this card. A VDI host bought now therefore moves to a later vGPU branch within its first year; our article on NVIDIA driver branches and CUDA versions explains the choice. NVIDIA's support matrix for vSphere requires a vSphere Foundation or Enterprise Plus licence for the card. Open-source serving engines on bare metal need no NVIDIA licence; NVIDIA AI Enterprise, licensed per GPU, is needed for NIM, so four cards need four licences, as our guide to NVIDIA AI Enterprise licensing explains.

Ordering in the EU, step by step

1. You describe the workload: models, desktops, 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, 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; 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 go 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 on your own chassis or with parts you already own, after a compatibility check of platform, power and cooling. Our GPU server acceptance testing guide lists checks for your own acceptance test on delivery, and our guide to GPU server quote comparison shows what each line of a quote should state.

Information that makes the first quote exact

A few sentences are enough to start. With these points we can quote the exact configuration in the first reply.

• Models with quantisation and context length, or desktop profiles and seats per user type
• Peak requests in flight, or the number of users and how they work
• MIG plans: which workloads need an isolated half
• Platform: bare metal, Kubernetes, vSphere or RHEL KVM, with the version
• Rack position: free height units, feed type and current, A/B feeds, PDU outlets, inlet temperature
• Whether the server should grow to eight cards later
• Delivery site, invoicing entity, timeline and any tender requirements, for which our GPU server tender specification helps

Supplied by us with the server

We supply the four-card server with NVIDIA RTX PRO 4500 Blackwell Server Edition cards in 2U chassis whose makers list four, and the eight-card version on our page to buy an 8× RTX PRO 4500 Server Edition server. The cards are also available on their own from our professional GPU range for servers you already run, and NVIDIA AI Enterprise and vGPU licences come on the same EU contract and invoice. If you are still choosing between the server card and the workstation card, our comparison of the RTX PRO 4500 or the RTX PRO 4500 Server Edition sets out the differences.

FAQ