GeForce in a server: the driver licence data centre clause and when RTX PRO is required

Can GeForce cards run in a data centre server

NVIDIA’s GeForce driver licence restricts data centre use of GeForce cards such as the RTX 5090. Section 2.8 of the NVIDIA Driver License Agreement, version of 25 February 2025, says that GeForce or Titan software “is not licensed for datacenter deployment”. The clause names GeForce and Titan only and does not mention RTX PRO or data-centre products. The hardware differs as well, since the RTX PRO 6000 Server Edition is a passive card specified for server airflow, with MIG and vGPU, while the GeForce RTX 5090 is a desktop card with its own cooler and neither feature.

The agreement does not define data centre deployment, and whether a particular installation falls under the clause is a legal assessment for the company’s legal department. This article quotes the licence and warranty texts as NVIDIA published them when we read them on 10 October 2026, and compares the cards on the points that matter in a server chassis. Speed on language models and training is covered in our RTX PRO 6000 vs RTX 5090 comparison.

What NVIDIA’s driver licence says

The agreement governs “the NVIDIA driver, and any additional software and materials provided”, and it binds the user “whether an individual or entity”. Section 1.1(c) grants a general right to “Deploy, for your own use, the SOFTWARE on infrastructure you own or lease”. Section 2.8 then narrows the grant for one product family. GeForce or Titan software “is licensed for use only on GeForce or Titan hardware products you own” and “is not licensed for datacenter deployment”. The term “datacenter” appears in section 2.8 and nowhere else in the agreement.

Section 2.7 is not limited to GeForce. Except as the agreement expressly grants, it rules out selling, renting, sublicensing, distributing or transferring the software, and it rules out providing “commercial hosting services with the SOFTWARE”. A company that runs GPU services for customers or other third parties reads section 2.7 as well as section 2.8, and checks which licence terms come with the drivers for the cards it buys.

The data centre clause is older than this agreement. The Register reported on 3 January 2018 that NVIDIA had added a section headed “No Datacenter Deployment” to its GeForce and Titan licence, with the exception that “blockchain processing in a datacenter is permitted”. The agreement of 25 February 2025 contains no such exception, so articles from 2018 that cite it describe the earlier text.

GeForce warranty and RTX PRO warranty

NVIDIA’s warranty for its own GeForce graphics cards, the Founders Edition models, runs three years from the date of purchase. It states that the product “is intended for consumer end user purposes only”, and it excludes data-centre use and GPU cluster commercial deployments, which it calls Enterprise Use. The next sentence reads “Any use of Warranted Product for Enterprise Use shall void this warranty.” The page gives 19 July 2016 as the date of the last update. GeForce cards bought from a board partner are, in NVIDIA’s words, “warranted by the partner”, so that partner’s terms apply instead.

NVIDIA’s warranty for RTX PRO workstation graphics cards also runs three years and names the RTX PRO 6000 Blackwell Workstation Edition and the Max-Q Workstation Edition. We found no restriction by place of use in it. Its exclusions concern defects and damage, for example problems “caused by abuse, misuse, negligence”, and it covers cards bought from NVIDIA or from a seller NVIDIA has authorised.

The Server Edition is not listed on that page, and NVIDIA’s warranty overview publishes no terms for data-centre GPUs; for most NVIDIA products it names the local distributor or reseller as the first contact for warranty and support. Server manufacturers state the warranty for the card as an option part in their own guides, as our article on servers compatible with the Server Edition shows. Our quotes state the warranty terms in writing, with manufacturer warranty handled through us.

RTX 5090 vs RTX PRO 6000 Server Edition and Max-Q

The three cards share the Blackwell generation, and the differences that decide a server build are in cooling, power delivery, partitioning and support lists rather than in raw speed.

NVIDIA product pages for the RTX 5090, the RTX PRO 6000 Server Edition and the Max-Q; RTX PRO 6000 Server Edition product brief SP-12355-001_v02 (June 2025); MIG supported GPUs (11 September 2026), vGPU supported GPUs (2 October 2026), AI Enterprise 8.2 support matrix (2 September 2026); NVIDIA Driver License Agreement (25 February 2025). All read on 10 October 2026.

The Max-Q and the Workstation Edition are built for workstations, where NVIDIA’s page for the Max-Q speaks of “up to four GPUs in a single system”. In a rack server the Server Edition is the edition with a server airflow specification and vGPU, as our comparison of the Workstation, Max-Q and Server editions explains in detail.

Cooling, airflow and power in a server chassis

NVIDIA’s product brief describes the Server Edition as “passively cooled” for “data center and edge server deployments”. Its airflow table asks for 43 CFM through the heat sink at a 25 °C heat sink inlet, rising to 51 CFM at 30 °C, 65 CFM at 35 °C, 84 CFM at 40 °C and 125 CFM at 45 °C. The figures assume fully ducted airflow. The 43 CFM minimum applies “whenever input power is provided to the card”, so the server must keep that airflow even at idle. The brief gives an ambient operating range of 10 to 45 °C.

The RTX 5090 is specified for a desktop system. NVIDIA’s product page gives 575 W of total graphics power, a required system power of 1,000 W for its reference configuration, and power through either one 600 W PCIe Gen 5 cable or four 8-pin cables with the adapter in the box. NVIDIA publishes no server airflow table for it.

The Server Edition draws its power through one 16-pin auxiliary connector, which the brief calls 12VHPWR. The brief sets the default limit by how the cable is strapped, 600 W with a cable strapped for 600 W and 450 W with one strapped for 450 W, and allows a power cap down to 300 W in both modes. Check which cable the server vendor lists for the slot and put it in the order with the card. The Server Edition also ships with its display outputs off by default and has a root of trust inside the GPU with secure boot, secure firmware upgrade and firmware rollback protection.

We supply the RTX PRO 6000 Server Edition as cards or in AI servers built to order, and we check the rack, power and airflow before we quote. Tell us the server model and how many cards it should hold.

Virtualisation, MIG and enterprise software support

NVIDIA’s list of MIG-capable GPUs, updated on 11 September 2026, includes all three RTX PRO 6000 editions with up to four instances, the RTX PRO 5000 and 4500 Blackwell with two, and the H200 NVL with seven, and it includes no GeForce card. The list of GPUs supported by vGPU, updated on 2 October 2026, names the RTX PRO 6000 Server Edition in its air-cooled and liquid-cooled versions, the RTX PRO 4500 Server Edition, the RTX PRO 5000 Workstation Edition with 72 GB, the L4 and the L40S, and no GeForce card. Our guide to MIG and vGPU shows how many virtual machines each profile allows.

NVIDIA’s support article on GeForce passthrough, updated on 29 September 2021, describes a beta that “is supported on R465 or higher drivers”. It passes one GeForce card to a single guest on a Linux host and states that “GeForce GPU passthrough supports 1 virtual machine” and that “SR-IOV is not supported on GeForce”. For several virtual machines on one card, NVIDIA points to “NVIDIA Tesla, Quadro, or RTX enterprise GPUs”.

The NVIDIA AI Enterprise 8.2 support matrix, updated on 2 September 2026, lists the RTX PRO 6000 and RTX PRO 4500 Server Editions, the L4, the L40S and the H200 NVL. It adds that these GPUs “are supported with servers listed on the NVIDIA-Certified Systems”, which ties enterprise software support to the server model as well as the card. On the driver side, the Server Edition brief names drivers from R575 on, and our guide to NVIDIA driver branches covers which production branch to run.

Which card for which use

The table maps common uses to the cards that fit them on the points above. It is a technical reading of NVIDIA’s published texts and lists, not a legal classification.

Licence sections from the NVIDIA Driver License Agreement (25 February 2025); MIG, vGPU and AI Enterprise lists and the Server Edition product brief as in the first table; H200 NVL memory and MIG count from NVIDIA’s MIG supported GPUs list.

For a GeForce card on a developer’s desk, the licence and warranty texts quoted above are the reference, and our 5090 comparison shows where the card is fast enough for one person. When the same card is to move into a rack or run inside virtual machines, the technical side changes as well, because NVIDIA publishes no server airflow figures for it and does not list it for MIG, vGPU or NVIDIA AI Enterprise.

If a GeForce prototype is moving into a server room, describe the model, the users and the target server in the form below, and we reply with a configuration and quote within one business day.

What we supply

We supply the RTX PRO 6000 Blackwell in all three editions, the RTX PRO 5000, 4500, 4000 and 2000, the L40S, the L4 and the H200 NVL as professional NVIDIA GPUs, with manufacturer warranty, on one EU contract and invoice. We do not supply GeForce cards. For the rack we build AI servers to order around the workload, assembled and burn-in tested, and we check the rack, power and airflow before we quote. NVIDIA AI Enterprise and vGPU licences come on the same invoice as the hardware.

FAQ