Buy a workstation with NVIDIA RTX 5880, 5000, 4500 or 4000 Ada Generation cards in Europe

Who still buys an RTX Ada workstation

You can buy a workstation with NVIDIA RTX 5880, RTX 5000, RTX 4500 or RTX 4000 Ada Generation cards from us, with one to four cards, built to order and delivered anywhere in the EU; the cards are available to order. These are the previous generation of NVIDIA's professional desktop cards, and for a new AI workstation without other constraints the RTX PRO Blackwell cards are the better choice, as the comparison further down shows. Our team sizes the configuration to your workload and budget and replies within one business day.

Companies order Ada cards today for three reasons:

• An existing Ada fleet. A design office or a lab that runs Ada workstations with one validated image, driver and set of containers adds the same card and keeps one configuration. Container images built only for compute capability 8.9 keep running.
• Certified applications. Where a CAD, simulation or visualisation vendor certifies specific cards, its list decides. If that list names an Ada card and not yet the Blackwell card you would buy instead, the Ada card keeps the seat inside the vendor's support.
• vGPU on a workstation-class card. NVIDIA lists the RTX 5000 Ada and the RTX 5880 Ada for vGPU, including on VMware vSphere. Of the RTX PRO Blackwell workstation cards only the RTX PRO 5000 72 GB is on the list, and on Red Hat KVM only.

For the 48 GB flagship, the RTX 6000 Ada, our page to buy an RTX 6000 Ada workstation covers the same questions.

The four cards side by side

NVIDIA datasheets for the RTX 5000 Ada (2898211, August 2023), RTX 4500 Ada (2895216, October 2023) and RTX 4000 Ada (2895141, September 2023), NVIDIA's RTX 5880 Ada product page and NVIDIA's vGPU supported GPU list (updated 2 October 2026), all read on 24 October 2026; RTX 5880 Ada bandwidth, bus and CUDA cores from its datasheet as cited in our RTX 5880 Ada comparison.

All four are actively cooled cards of 4.4 inches height, PCIe 4.0 x16, with four DisplayPort 1.4a outputs and AV1 encoding. The RTX 4000 Ada is 9.5 inches long and takes one slot; the other three are 10.5 inches long and take two. Their Tensor Cores support FP8 but not FP4, and NVIDIA's MIG user guide lists no Ada GPU, so none of them splits into isolated instances. NVIDIA's datasheets state no NVLink for the RTX 5000, 4500 and 4000 Ada, and our comparison of the RTX 5880 Ada and RTX 6000 Ada records the same for the 5880.

When Blackwell is the better buy

Each Ada card except the RTX 5880 Ada has an RTX PRO Blackwell counterpart at the same tier, as our tier-by-tier comparison of Ada and Blackwell lists.

NVIDIA datasheets and product pages of each card, read on 24 October 2026 and as cited in our Ada and Blackwell comparison.

On language models the bandwidth column decides the speed per user, because each generated token reads the weights once. The RTX PRO 4000 has 87 per cent more bandwidth than the RTX 4000 Ada in the same single slot, the RTX PRO 4500 107 per cent more than the RTX 4500 Ada at 10 W less, and the RTX PRO 5000 133 per cent more than the RTX 5000 Ada. The Blackwell cards add FP4 and PCIe 5.0, and the RTX PRO 5000 and 6000 add MIG.

The RTX 5880 Ada sits closest to the RTX PRO 5000 48 GB: the same 48 GB, 960 against 1,344 GB/s, 285 against 300 W, and MIG on the Blackwell card. The 5880 keeps its place where vGPU on vSphere is needed, which the RTX PRO 5000 48 GB lacks. For the RTX 4500 Ada and the RTX 4000 Ada, which are not on NVIDIA's vGPU list, the only reasons left are an existing fleet and a certification list; otherwise our pages to buy an RTX PRO 4000 workstation and to buy an RTX PRO 4500 workstation show the Blackwell builds. We supply both generations; send us the cards your teams run today and the applications on them, and our team replies with the Ada card or its Blackwell counterpart for each seat.

Workstations that offer four Ada cards

Dell Precision 7960 configurator (Canada), Lenovo PSREF ThinkStation PX specification of 9 October 2026 and HP Z8 Fury G6i specifications, read on 24 October 2026.

Dell's configurator offers the Precision 7960 with one, two, three or four cards of each model: "Quad Nvidia RTX 5880 Ada", "Quad NVIDIA RTX 5000 Ada Generation", "Quad NVIDIA RTX 4500 Ada Generation" and "Quad NVIDIA RTX 4000 Ada Generation". Lenovo lists the RTX 5000, 4500 and 4000 Ada for the ThinkStation PX with their power and slot width, but gives a maximum of four only for the RTX 6000 Ada and the Max-Q, and does not list the RTX 5880 Ada. HP's current Z8 Fury G6i page lists no Ada card. Because all four Ada cards are PCIe 4.0, the two PCIe 4.0 x16 slots that Dell lists for the Precision 7960, next to two PCIe 5.0 x16 slots, cost them nothing.

We build four-card Ada workstations to order on Threadripper PRO or Xeon W platforms, or supply the cards for one of these towers after a check of slots, power supply and cables.

What we put in the workstation

Our AI servers page (platforms and components); memory and cores by NVIDIA's configuration guide for NVIDIA-Certified Systems (30 September 2026), written for servers and applied here by the same arithmetic: at least 6 cores per GPU and twice the GPU memory.

By that rule four RTX 5880 Ada need 384 GB of system memory, four RTX 5000 Ada 256 GB, four RTX 4500 Ada 192 GB and four RTX 4000 Ada at least 160 GB. Each card takes 16 PCIe 4.0 lanes, 64 for four.

What one to four Ada cards run

The counts follow our users-per-card rule: 90 per cent of the memory the driver reports, less 3 GiB per card, less the weights, divided by the FP8 KV cache of one conversation of 8,192 tokens, 0.5625 GiB for Qwen3-8B, 0.625 GiB for Qwen3-14B, 1.0 GiB for Qwen3-32B and 1.25 GiB for Llama 3.3 70B. GDDR6 cards lose 6.25 per cent of their memory with ECC enabled, as NVIDIA's CUDA Best Practices Guide states, so the tables give both cases. We scale the driver-reported memory of a 48 GB Ada card, 44.8 GiB with ECC on and 47.8 GiB with ECC off, by capacity for the smaller cards, an assumption to check with nvidia-smi -q.

Our estimates by the users-per-card rule, concurrent conversations of 8K with an FP8 KV cache on one card, ECC on and ECC off in brackets; weights Qwen3-8B FP8 8.8 GiB, Qwen3-14B FP8 15.2 GiB. Memory arithmetic, not measured throughput.

Our estimates by the users-per-card rule, ECC on and ECC off in brackets; Qwen3-32B FP8 32.0 GiB, Llama 3.3 70B FP8 67.7 GiB; the RTX 5880 Ada row equals our RTX 6000 Ada figures, since both cards have 48 GB.

FP8 is the precision to plan with on Ada: the cards have FP8 Tensor Cores and no FP4, so NVFP4 checkpoints belong on Blackwell, and for other 4-bit formats the serving engine's support on Ada has to be checked first. Models up to 8B run one copy per card on every Ada card here; a 14B model fits from the RTX 4500 Ada, but leaves room for more than a handful of conversations only from the RTX 5000 Ada upwards. A 70B model in FP8 needs four RTX 5000 Ada or four RTX 5880 Ada to leave room for conversations. Splitting over four cards runs over PCIe 4.0, and by default vLLM switches its custom all-reduce off on more than two PCIe-only GPUs, so TP=4 runs on NCCL; vLLM 0.30.0 (release notes, 22 September 2026) added an opt-in FlashInfer PCIe IPC all-reduce for systems without NVLink (#53576).

Bandwidth sets the speed per user. By our arithmetic, Qwen3-14B in FP8 reads about 16.3 GB per token, which limits one conversation to about 59 tokens per second on an RTX 5880 Ada, 35 on an RTX 5000 Ada, 27 on an RTX 4500 Ada and 22 on an RTX 4000 Ada, against 41 on an RTX PRO 4000. These are ceilings from the specifications, not measurements. Our comparison of 48 GB GPUs: RTX 6000 Ada, RTX 5880 Ada, RTX PRO 5000 and L40S sets the 48 GB cards side by side.

vGPU, display outputs and drivers

NVIDIA's list of GPUs supported by vGPU, updated on 2 October 2026, includes the RTX 5000 Ada from release 16.1 and the RTX 5880 Ada from 17.0, both with full support and no end date set, and does not include the RTX 4500 Ada or the RTX 4000 Ada. NVIDIA's product pages for the RTX 5000 and RTX 5880 Ada list NVIDIA vPC/vApps and NVIDIA RTX Virtual Workstation and add that the display ports are on by default: "Turn display ports off when using vGPU software." NVIDIA's support matrices of 29 September 2026 list the RTX 5000 Ada for vSphere 8.0, VCF 9.0 and 9.1 and for Red Hat Enterprise Linux KVM, as our RTX PRO 5000 and RTX 5000 Ada comparison records, and the RTX 5880 Ada for VCF 9.1, VCF 9.0 and ESXi 8.0. Without MIG, virtual machines share an Ada card in time slices; our article on GPUs in VMware vSphere: passthrough or vGPU covers the host side.

NVIDIA's data-centre drivers 580.178.04 and 595.91.07 of 3 August 2026 list the RTX 5880, 5000, 4500 and 4000 Ada beside the RTX PRO Blackwell cards, so one installed driver covers both generations, as our guide to mixing Ada and Blackwell GPUs explains. Ada support arrived with CUDA 11.8, and the cards have compute capability 8.9. On Linux NVIDIA recommends its open GPU kernel modules for Ada and requires them for Blackwell, so a workstation that may later take a Blackwell card is best set up on the open modules from the start.

Power on 230 V

Our estimate at full load from NVIDIA's board power and a 350 W workstation processor such as the AMD Threadripper PRO 9975WX (32 cores, default TDP 350 W), plus 15 per cent for memory, drives, fans and power-supply losses; currents at 230 V. A 16 A circuit planned at 80 per cent carries 2.94 kW.

Every four-card build fits one 230 V 16 A circuit with room to spare. The heat output equals these figures: about 1.7 kW from four RTX 5880 Ada under continuous load calls for a technical room or an air-conditioned office, while four RTX 4000 Ada put about 1.0 kW into the room. Large workstation supplies reach their full rating at 200 V or more; Dell offers the Precision 7960 with a 1,400 W or a 2,200 W chassis, and the configurator decides which one a given card count needs.

Software on delivery

On request the workstation arrives with your operating system or company image, the NVIDIA driver branch your Ada fleet runs, CUDA and a container runtime with the NVIDIA Container Toolkit. As of October 2026 the R580 branch has long-term support to June 2028, and NVIDIA's security bulletin of 30 September 2026 names 580.178.04 and 595.91.07 as fixed versions; our guide to NVIDIA driver branches and CUDA versions covers pinning the branch. For vGPU hosts with the RTX 5000 or 5880 Ada, the vGPU software and licences come in the same quote.

Ordering in the EU

1. You describe the workload and the fleet: cards and drivers in use, applications and their certification, users, room and timeline. Sizing is free of charge.
2. We propose the configuration, with a Blackwell alternative where it fits, and check cards, tower, power supply and cooling.
3. You receive a written quote within one business day, with 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 workstation 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 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.

What to tell us for a quote

• The Ada cards, driver branch and images your teams run today
• The applications and whether their vendor certifies specific cards
• Models and users, or the vGPU profiles and hypervisor if virtual machines are planned
• The tower your company standardises on, if any
• Circuit rating, cooling and where the workstation will stand
• Delivery site, invoicing entity and timeline

List these points in the form below, and we reply with a configuration and a written quote within one business day, with a Blackwell option alongside where it suits the work.

FAQ