NVIDIA RTX cards in volume for system builders and IT departments, from 4 GB to 48 GB

NVIDIA RTX cards in volume, and who orders them from us

Eurokommerz supplies NVIDIA professional graphics cards across the EU to system builders, integrators and IT departments. This page covers eleven of them, from the RTX A400 with 4 GB to the RTX PRO 5000 Blackwell with 48 GB; the cards above 48 GB are on our GPU page. A system builder that fits one card type into a series of workstations, an integrator that equips control rooms or compact edge systems, and an IT department that renews its CAD seats all get the same from us: a written quote within one business day, one European contract and an invoice from Austria.

Companies that search for an NVIDIA distributor, an NVIDIA RTX reseller or a GPU wholesaler usually need a GPU supplier in Europe for a defined card in a defined quantity, with the paperwork in one place. This page is written for that case. It lists the cards, says what each needs from the system it goes into, and names what an enquiry should contain so that it can be quoted without a second round. How the buying channels for GPU hardware differ is set out in our guide on where to buy an AI server in Europe. For IT resellers and system houses that sell the cards on to their own customers, NVIDIA's partner channel is described in our guide to NVIDIA GPUs for resellers in Europe.

Eleven cards from 4 GB to 48 GB and the fleets they go into

NVIDIA datasheets RTX A400 (3279761, May 2024), RTX A1000 (April 2024), RTX 2000E Ada (4328932, October 2025), RTX 2000 Ada (2999417, January 2024), RTX PRO 2000 (5322351, June 2026), RTX PRO 4000 (5323450, June 2026), RTX 4500 Ada (2895216, October 2023), RTX PRO 4500 Workstation Edition (5108623, April 2026), RTX PRO 4500 Server Edition (5192801, April 2026), RTX 5000 Ada (2898211, August 2023) and RTX PRO 5000 (5349550, June 2026). The fleet column is our mapping by memory, format and listed features. Each card name links to its buy page.

The eleven cards come from three generations: the RTX A400 and RTX A1000, which NVIDIA announced on 16 April 2024 on its Ampere architecture, the RTX Ada Generation, and RTX PRO Blackwell. NVIDIA lists ECC memory for the Ada and Blackwell cards and states none for the two RTX A cards. A series does not have to change generation because a newer one exists. A builder who has validated software and images on an Ada card has a reason to stay with it, and all eleven cards are available to order. Our guide to NVIDIA RTX workstation GPUs from 4 to 48 GB sets out how the generations relate.

The complete figures are on the spec pages: RTX A400, RTX A1000, RTX 2000E Ada, RTX 2000 Ada, RTX PRO 2000 Blackwell, RTX PRO 4000 Blackwell, RTX 4500 Ada, RTX PRO 4500 Workstation Edition, RTX PRO 4500 Server Edition, RTX 5000 Ada and RTX PRO 5000 with 48 GB.

Put the exact product into the order

A product name alone can point to two different PCIe cards. In a volume order that is worth one more line in the request.

NVIDIA datasheets 5323450 (June 2026), 4016700 (August 2025), 5108623 and 5192801 (April 2026), 5349550 (June 2026), 2999417 (January 2024), 4328932 (October 2025) and the RTX A1000 datasheet (April 2024); NVIDIA's laptop line card of February 2024.

Laptop GPUs share their names with several of these cards: RTX PRO 2000, 4000 and 5000 Blackwell Generation, RTX 2000 Ada and RTX 5000 Ada, and RTX A1000. They are components of mobile workstations with other memory sizes and cannot be ordered as cards, but their figures turn up in listings and benchmark tables under the same name. The reverse mistake also occurs: NVIDIA lists no Server Edition and no Max-Q version of the RTX PRO 5000, 4000 or 2000, as our article RTX PRO 5000 Server Edition: which cards exist explains.

We supply all eleven cards of this page across the EU; send us the exact card, the quantity and the systems it goes into and we reply with a written quote within one business day.

What the system has to provide for each class of card

Mechanically the eleven cards fall into two classes, and a chassis series is usually built for one of them.

• Half height, 50 to 70 W. The RTX A400, RTX A1000, RTX 2000E Ada, RTX 2000 Ada and RTX PRO 2000 Blackwell are 2.7 in high, the format usually called low profile. The first three are single slot, the last two dual slot. NVIDIA's quick start guides say that the RTX 2000 Ada and the RTX PRO 2000 do not use a power cable. For the RTX A1000 and RTX A400, NVIDIA's datasheet gives a maximum power consumption of 50 W and names no auxiliary power connector; the RTX 2000E Ada datasheet gives 50 W of total board power and has no connector row. All five use an x8 link on what NVIDIA calls a full-length PCIe interface, PCIe 4.0 on the Ada and RTX A cards and PCIe 5.0 on the RTX PRO 2000.
• Full height, 145 to 300 W. The RTX PRO 4000 Blackwell and the RTX PRO 4500 Server Edition are single slot; the RTX PRO 4500 Workstation Edition, the RTX PRO 5000, the RTX 5000 Ada and the RTX 4500 Ada are dual slot. All six take the connector NVIDIA calls PCIe CEM5 16-pin. By NVIDIA's quick start guides the boxes of the five workstation cards contain an adapter, from one 8-pin PCIe cable for the RTX PRO 4000 and from two for the other four. The RTX PRO 4000 is 9.5 in long, the others 10.5 in.
• The passive card in the second class. The RTX PRO 4500 Server Edition is cooled passively. Its heat sink accepts air from left to right or from right to left, NVIDIA's product brief gives an ambient range of 10 °C to 45 °C, and NVIDIA publishes no required airflow figure. The airflow has to come from the server chassis.

NVIDIA publishes no power-supply wattage for any of the eleven. The check for a series is therefore arithmetic on the system maker's figures: the board power of the cards plus the load of the rest, against the rating of the supply. For the Blackwell workstation cards NVIDIA's quick start guide also asks for at least as much system memory as GPU memory and recommends twice as much, which is 96 GB for a 48 GB card by the recommendation.

Three of our articles go into these points card by card: RTX power connectors from the A400 to the RTX PRO 5000, low-profile NVIDIA RTX cards compared and single-slot NVIDIA RTX cards compared.

One driver package for a mixed fleet

A builder that ships several card types wants one image. For the ten workstation cards NVIDIA's release notes allow that. The RTX Enterprise driver 621.14 of 4 November 2026 lists all ten, and so do 597.30 and 582.99 in the two older branches. Windows 11 and Windows 10 from version 22H2 are named as tested systems, and Release 620 is the last branch for Windows 10. The Server Edition is the exception with the data-centre driver, R595 or later. Versions, end dates and the Linux releases are in our guide to the NVIDIA RTX Enterprise driver on Windows 10 and 11.

Validation is done per system. In Dassault's SOLIDWORKS 2026 certification list, read 10 November 2026, a row certifies one computer model with one card and one driver. Rows with desktop workstations exist for the RTX A400 and A1000, the RTX 2000, 4500 and 5000 Ada and the RTX PRO 2000, 4000, 4500 and 5000 Blackwell; the list separates neither the two RTX PRO 4500 cards nor the two memory sizes of the RTX PRO 5000, and it has no entry for the RTX 2000E Ada. A builder of CAD workstations checks the row for its own model or for the platform it builds on.

Product life and driver coverage for a multi-year series

A series that is built over two or three years raises the question of how long the same card can be ordered. NVIDIA answers it for one product family only: its long-life availability statement covers RTX PRO embedded modules. The "RTX A1000" and the "RTX 2000 Ada Generation" on that page are modules with 4 GB and 8 GB, not the desktop cards of this page, and the RTX 2000E Ada is not on it either, although NVIDIA's datasheet describes that card as built for embedded and edge computing. For the eleven cards of this page NVIDIA publishes no long-life statement, so a series that runs over several years is planned order by order.

For drivers NVIDIA does publish a date. All eleven cards are listed in driver branch R620, which NVIDIA's data-centre driver table of 4 November 2026 supports to September 2029.

How a volume enquiry runs with us

1. Enquiry. You send the card under its exact name, the quantity, the delivery schedule you need and the destination.
2. Quote. We reply within one business day with a written quote; quantities and lead time are confirmed in writing with the quote.
3. Contract and invoice. The order runs on one European contract, and the invoice comes from Austria.
4. Delivery. We deliver anywhere in the EU. Other destinations go through our logistics partner.
5. Warranty. The manufacturer warranty is handled through us, and a card that arrives dead is replaced.

An enquiry that can be quoted at once contains five things: the card, with its edition, memory size or board power as in the table above; the quantity, in total and per delivery if the cards should arrive in several lots; the delivery schedule your production or rollout plan sets; the destination, meaning the delivery address and the company that receives the invoice; and the systems the cards go into, so that the quote names the version that matches the chassis. What NVIDIA itself publishes for its professional GPUs, and what a written quote should contain, is set out in our guide to requesting NVIDIA GPU quotes in Europe.

If the card for a series is not fixed yet, tell us the application, the chassis and the number of systems. The first consultation is free of charge.

Cards alone or built into workstations and servers

Not every IT department wants to fit cards itself. We also supply workstations and servers built to order with these cards; a card built into a system is then a separate line of the system quote. For builders and integrators the usual order remains the card on its own, and the pages linked in the first table describe each one.

FAQ