Buy the NVIDIA RTX 5000 Ada Generation 32 GB: price on request, supplied across the EU

RTX 5000 Ada 32 GB price on request and supply in the EU

The NVIDIA RTX 5000 Ada Generation is a dual-slot workstation graphics card of the Ada Lovelace generation with 32 GB of GDDR6 memory with ECC, 12,800 CUDA cores and 250 W of total board power. We supply it to companies across the EU. Its price is not printed on this page: it reaches you as a written quote within one business day, together with the lead time and the warranty terms.

NVIDIA announced the card on 8 August 2023 and the first RTX PRO Blackwell cards on 18 March 2025. A buyer in late 2026 therefore checks first whether an Ada card is still the right order. For a number of companies it is, for reasons that can be named exactly, and for others the Blackwell card with the same 32 GB is the better buy. The complete figures are in our RTX 5000 Ada specs.

Why companies still order the RTX 5000 Ada in late 2026

1. A fleet that already runs it. Seats added to a fleet of RTX 5000 Ada workstations get the same part with the same driver and the same power adapter. The card is listed in the newest release of each current driver branch, R580, R595 and R620. NVIDIA's data-centre driver table of 4 November 2026 lists R620 as a Long Term Support Branch until September 2029; for the Windows RTX Enterprise branches NVIDIA publishes no end date.
2. An image qualified on Ada. NVIDIA's CUDA page gives the card compute capability 8.9 and the RTX PRO Blackwell cards 12.0. A company that has validated its image, its driver and its applications on the Ada card keeps that validation when the next card is the same model.
3. Virtual workstations. NVIDIA's vGPU supported-GPU list of 2 October 2026 includes the RTX 5000 Ada from release 16.1. The NVIDIA RTX PRO 4500 Blackwell Workstation Edition and the NVIDIA RTX PRO 5000 Blackwell with 48 GB are not on that list.
4. FP8 software. NVIDIA's datasheet states the Tensor performance of the card in FP8, and its NIM support matrix of 20 October 2026 ties FP8 to compute capability 8.9. A stack built on FP8 engines has what it needs on this card. FP4, which the Blackwell cards add, is not accelerated in hardware on Ada.
5. Published results. SPEC's result pages hold runs of systems with this card in SPECviewperf 15.0 and in SPECapc for SolidWorks 2024, which helps a buyer who has to document expected performance. For the Blackwell cards we found no SPEC result.

None of the five reasons is about speed; where none applies, the last section names the card to order instead. We supply the RTX 5000 Ada across the EU; send us the quantity and the workstation models and we reply with a written quote within one business day.

The quote, the price factors and what to send

You receive the quote in writing within one business day of your request. It states the price for your quantity, the lead time and the warranty terms; quantities and lead time are confirmed in writing with the quote. The card is available to order.

Three things move the price: the quantity, the delivery address and the date of the order. We therefore publish no list price and no MSRP figure. A card that goes into a workstation built to order is one line in a quote for the whole system, as our page on workstations and servers built to order describes; the page to buy an RTX Ada workstation shows such builds.

A complete request names the number of cards, the desktop card with 32 GB, the workstation models and the delivery address with the invoicing company.

Desktop card with 32 GB, laptop GPU with 16 GB

Searches for an RTX 5000 Ada with 16 GB lead to a different product, because NVIDIA uses the name three times.

NVIDIA RTX 5000 Ada Generation datasheet 2898211 (August 2023); NVIDIA's "Compare Professional Laptop GPUs" page, as read on 10 November 2026.

Only the first row is a PCIe card. A third product, an embedded GPU of the same name, also has 16 GB. NVIDIA announced the laptop GPU on 21 March 2023, before the desktop card, and driver lists write the other two as "Laptop GPU" and "Embedded GPU". Certification rows and benchmark results for mobile workstations appear under nearly the same card name, in Dassault's SOLIDWORKS list as on SPEC's result pages, which is why the system model in each row matters. In a request to us the words "desktop card, 32 GB" settle the matter.

The card in a buyer's checklist

NVIDIA RTX 5000 Ada Generation datasheet 2898211 (August 2023); NVIDIA's quick start guide for the RTX Ada cards (2023); Video Encode and Decode GPU Support Matrix of 16 October 2026; vGPU supported-GPU list (2 October 2026); MIG User Guide (4 November 2026); RTX Enterprise release notes 621.14, 597.30 and 582.99.

The datasheet lists a third display line, four displays at 5120 × 2880 and 60 Hz; the lines are alternatives, and four displays is the most one card drives. NVIDIA gives the size in inches, and converted it is about 112 by 267 mm. It publishes no millimetre figure, no weight and no fan type.

For local AI models, our sizing rule gives the card a budget of 25.7 GiB for model weights plus KV cache. That is 0.9 times the memory the driver reports, less 3 GiB; we derive the reported memory from the 95.6 GiB the driver shows on the 96 GB card, an assumption that the driver on the delivered card settles. By file size on Hugging Face, read 10 November 2026, Gemma 4 12B-it in BF16 (22.28 GiB) and Devstral Small 2 24B Instruct in FP8 (24.02 GiB) lie inside that budget. Which other files fit is worked out in our article on AI on the RTX 5000 Ada and RTX 4500 Ada.

Slot, power cable, adapters and driver

The card is 4.4 in high and 10.5 in long, takes two slots and needs a PCIe x16 slot. NVIDIA's datasheet has no row for the power connector. The quick start guide for the Ada cards describes a 16-pin PCIe CEM5 connector on the back edge of the card and lists an adapter from two 8-pin PCIe cables in the box. It tells the installer to connect two separate 8-pin cables from the power supply and to use only the adapter that comes with the card.

A power-supply wattage is not among NVIDIA's figures for the card. What can be planned is the 250 W of board power per card, set against the rated output of the workstation's power supply. The card has no HDMI output, and the quick start guide lists one DisplayPort-to-HDMI adapter in the box.

The driver is the NVIDIA RTX Enterprise Production Branch driver, and all three current Windows releases list the card: 621.14 in R620 (4 November 2026), 597.30 in R595 (22 October 2026) and 582.99 in R580 (21 October 2026). Their release notes cover Windows 11 from 22H2 to 25H2 and still name Windows 10 22H2, for which Release 620 is the last branch. For Linux they name, among others, the Ubuntu LTS releases 22.04 to 26.04, RHEL and Rocky from 8.y to 10.x, SLES 15 SP7 and 16, and Debian 12 and 13.

vGPU on the RTX 5000 Ada as NVIDIA documents it

NVIDIA's vGPU supported-GPU list gives the RTX 5000 Ada full support from release 16.1, and the datasheet names NVIDIA vPC, vApps and NVIDIA RTX Virtual Workstation as the software. A datasheet footnote adds that the display ports are on by default and have to be turned off when vGPU software is used. A card that serves virtual desktops therefore drives no local monitor.

As of November 2026 NVIDIA's vGPU index lists release 21.0 on the R620 driver branch, 20.3 on R595 and 19.7 on R580, and the Virtual GPU Types Reference of release 21.0 (5 November 2026) has a section for the card. NVIDIA's vGPU documentation also states that in GPU pass-through deployments all GPUs that support vGPU software are supported.

MIG is a separate mechanism, and NVIDIA's MIG User Guide of 4 November 2026 lists no Ada card. Which cards can be shared, and how, is the subject of MIG and vGPU on RTX PRO and RTX Ada cards. If the cards are meant for a virtualisation host, tell us the hypervisor and the number of users, and we reply with a written quote for that host.

Ordering for a fleet, with contract, delivery and warranty

One European contract covers the order, and the invoice comes from Austria. Delivery goes anywhere in the EU; other destinations are handled through our logistics partner. The card carries the manufacturer warranty, which is handled through us, and units that arrive defective are replaced. The first consultation is free of charge.

IT departments and system builders who need larger quantities name the total and the systems, and the quote covers the whole quantity; our page on NVIDIA RTX cards in volume for system builders describes such orders. Fleets that still run RTX A4500 or RTX A5000 cards find the step to this card in RTX 4500 Ada and RTX 5000 Ada vs RTX A4500 and RTX A5000.

RTX PRO 4500 Blackwell or RTX PRO 5000 as the alternative

NVIDIA datasheets 2898211 (August 2023), 5108623 (April 2026), 5349550 (June 2026) and 2895216 (October 2023).

The RTX PRO 4500 Blackwell Workstation Edition has the same 32 GB with 56 per cent more memory bandwidth by our arithmetic, at 50 W less, in the same dual-slot size of 4.4 by 10.5 in. It adds FP4 on its Tensor Cores and 4:2:2 encoding, which NVIDIA's video support matrix does not list for Ada cards. Blender Open Data, read 10 November 2026, shows for Blender 4.5.0 a median of 8,486.57 samples per minute over 17 submissions for it and 5,677.13 over 9 submissions for the RTX 5000 Ada; the database is crowd-sourced. For a new seat without one of the five reasons above it is the card to order: buy the RTX PRO 4500 Blackwell, its 32 GB spec page and RTX 5000 Ada vs RTX PRO 4500 Blackwell.

The RTX PRO 5000 Blackwell with 48 GB is the step for work that outgrows 32 GB. In Puget Systems' professional GPU roundup of 18 December 2025, on a test bench with a Ryzen 9 9950X3D, the RTX PRO 5000 Blackwell came out 50 per cent ahead of the RTX 5000 Ada in Blender Cycles and 35 per cent ahead in V-Ray GPU. See the page to buy the RTX PRO 5000 Blackwell 48 GB, RTX PRO 5000 48 GB vs 72 GB and RTX PRO 5000 Blackwell vs RTX 5000 Ada.

Inside the Ada generation the NVIDIA RTX 4500 Ada Generation is the smaller order when 24 GB is enough: buy the RTX 4500 Ada, RTX 4500 Ada specs and RTX 4500 Ada vs RTX 5000 Ada. How the two generations line up tier by tier is in Blackwell vs Ada. If the decision between Ada and Blackwell is still open, describe the fleet and the applications, and we quote the card that fits, or both.

FAQ