NVIDIA vGPU licensing: RTX vWS, vPC and vApps editions, concurrent users and terms

NVIDIA vGPU licences in brief

NVIDIA licenses graphics vGPU in three editions, all per concurrent user. NVIDIA Virtual Applications (vApps) is for published applications and session-based desktops, NVIDIA Virtual PC (vPC) for office desktops, and NVIDIA RTX Virtual Workstation (RTX vWS) for CAD, 3D and GPU computing inside a desktop. Each comes as a subscription or as a perpetual licence with five years of support, upgrades and maintenance (SUMS).

This guide covers what each edition allows, how NVIDIA counts users, the terms on offer and what happens to a VM without a licence, from NVIDIA's packaging and licensing guide (updated 24 March 2026), its client licensing user guide (updated 29 September 2026) and its "buy" page for vGPU software (published 3 June 2026), all read on 24 October 2026. Prices are not part of this article; NVIDIA publishes list prices in its guide, and we quote the licences with the hardware. For licensing of compute vGPU and the AI software stack, see our guide to NVIDIA AI Enterprise licensing; for the licence server itself, our guide to DLS and CLS.

What vApps, vPC and RTX vWS each allow

NVIDIA's guide sets out the editions in one entitlement table. The rows that decide which edition a user needs are these.

NVIDIA Virtual GPU Software Packaging, Pricing, and Licensing Guide, product entitlement table and supported profile tables, last updated 24 March 2026, read 24 October 2026. The vApps display limit "Applies only to the console display in remote application environments". CCU: concurrent user.

Three points follow from the table. vPC is the desktop edition without GPU computing: an office user who runs a CUDA application in the desktop needs RTX vWS. The profile range on Blackwell starts at 2 GB for vPC, because NVIDIA offers no 1 GB profiles on Blackwell GPUs, so a desktop pool that ran 1 GB profiles on older cards moves to 2 GB on an RTX PRO Server Edition. And RTX vWS is the only edition that lets one VM use several vGPUs, which a render or simulation VM needs.

The editions also map onto the profile series. The client licensing guide assigns A-series vGPUs to vApps, B-series to vPC and Q-series to RTX vWS, and adds that "The vWS license edition includes the vPC license edition", so a B-series desktop can run on a vWS licence as well. Passthrough and bare-metal GPUs used for professional 3D graphics need vWS; used for PC-level applications they fall under vApps, enforced only through the end-user licence agreement.

How NVIDIA counts concurrent users

The guide defines the count in one sentence: "A CCU license is required for every user who is accessing or using the software at any given time". Its terminology section allocates licences by concurrently active VMs, one VM per licence, and session-based solutions by concurrent user. Three details from the client licensing guide decide how that works in practice.

1. A VM checks out its licence when it boots and returns it at shutdown. NVIDIA's setting for checkout at user login and release at logout changes that, and it matters for pools of desktops that stay powered on overnight.
2. A VM with several Q-series vGPUs needs "only one license for the VM", not one per vGPU.
3. For A-series vGPUs on a session host, "one vApps license is required for each concurrent user". Only one of those users is enforced in software; the rest are covered by the licence agreement.

A worked example shows the effect. A company has 400 office staff, of whom at most 300 are logged in to a virtual desktop at the same time, 40 CAD users with at most 25 sessions at once, and a published-application farm with 200 concurrent sessions. It needs 300 vPC, 25 RTX vWS and 200 vApps licences by NVIDIA's rule. If the 300 office desktops stay booted all night under the default checkout at boot, they keep holding 300 licences, which is still within the count. If 350 desktops are booted, 350 licences are checked out. The count to license is the peak of licences in use, and the licence server's usage reports show it.

NVIDIA's guide counts users, not GPUs, so the card in the host does not change the count. 300 office desktops need 300 vPC licences on an RTX PRO 4500 Server Edition host or an L4 host alike; the number of cards follows from the profile size, as our guide to VDI GPU sizing explains.

We supply RTX vWS, vPC and vApps licences together with the vGPU cards and servers, on one EU contract and invoice; send us your peak concurrent users per user type, and our team recommends the edition and term.

Subscription or perpetual: the terms NVIDIA offers

The packaging guide lists two models for each edition.

NVIDIA Virtual GPU Software Packaging, Pricing, and Licensing Guide, subscription and perpetual pricing tables (term descriptions only), last updated 24 March 2026, read 24 October 2026.

A subscription is "active for a fixed period" and has to be renewed at the end of the period to stay active; SUMS is included for its term. A perpetual licence "allows for use of the licensed software indefinitely", but NVIDIA states that "All NVIDIA vGPU software products with perpetual licenses must be purchased in conjunction with a five-year SUMS", and that "A one-year SUMS is available only for renewals". After the first five years, SUMS can be renewed year by year.

NVIDIA's buy page, published on 3 June 2026, differs from the guide in two places. It says that SUMS for perpetual licences "is available in four or five year increments", where the guide names five years. And its introduction says vGPU software can be bought per concurrent user "or per GPU as an annual subscription", without naming the editions or listing such an option; the guide itself describes no per-GPU model for vApps, vPC or RTX vWS. Have the quote state the SUMS term and the counting basis in writing.

SUMS is what keeps a perpetual licence current. NVIDIA's guide lists access to maintenance releases, defect fixes and security patches, new major releases, long-term branch maintenance for up to three years from general availability, and direct access to NVIDIA support. Support is at the Business Standard level by default, with Business Critical as an option. Without SUMS the licence keeps working, but newer vGPU releases, and with them support for newer GPUs, are no longer covered.

Which edition each user type needs

NVIDIA's guide ends with a short table for choosing the licence. With the profile series added, it reads as follows.

NVIDIA packaging and licensing guide, "Choosing the Right License" (24 March 2026); profile series from NVIDIA's client licensing user guide (29 September 2026); C-series from NVIDIA AI Enterprise documentation, as cited in our AI Enterprise licensing guide.

The guide's supported-GPU table assigns an edition range to each card. For the cards we supply it lists the RTX PRO 6000 Blackwell Server Edition for high-end RTX vWS and for vPC, the RTX PRO 4500 Blackwell Server Edition for entry RTX vWS and for vPC, the L40S for high-end RTX vWS, and the L4 for entry to mid-range RTX vWS and for vPC. Only the Server Editions of the RTX PRO cards are on NVIDIA's vGPU list; an RTX PRO 6000 Max-Q or Workstation Edition reaches a VM only by passthrough, which needs a vWS licence on Linux, as our article on the RTX PRO 6000 Max-Q in virtual machines explains.

The edge case is the user who needs both: a designer who also trains small models in the same VM. RTX vWS covers CUDA, OpenCL and AI development in a Q-series desktop. A VM that runs only compute on a C-series profile is outside all three editions and needs NVIDIA AI Enterprise, which NVIDIA licenses per GPU, so a server with both kinds of VM carries two licence lines.

What happens without a valid licence

The packaging guide states that "the vGPU software runs with or without a valid license server connection", so that the user experience is not impaired by licence overages or connection problems, and that the licence server lets administrators track usage. The client licensing guide describes what an unlicensed VM does in the software-enforced cases. When a VM boots without a licence, its vGPU or GPU "operates at full capability"; its performance "is degraded if the VM fails to obtain a license within 20 minutes", and further degraded if it still has none a day later. Where licensing is enforced only through the licence agreement, no licence is checked out, and nvidia-smi -q still reports the product as licensed.

In practice the licence server has to stay reachable for every VM that boots, and the count bought has to cover the peak: an overage does not stop users at once, but every concurrent user still needs a licence. The licence server's own options, cloud-hosted CLS or the on-premise DLS appliance, and how VMs receive their client configuration token are in our guide to the NVIDIA licence server.

Planning the licence count with the hardware

The licence lines follow from the same inputs as the hosts. Collect the peak concurrent sessions for each user type, give each type a profile, then derive the cards and the licences side by side.

• Count concurrent sessions at the busiest hour, not staff or named accounts, and check whether desktops stay booted outside working hours.
• Put CAD and 3D users on RTX vWS even if their profile is a B type, if they run CUDA or OpenCL applications.
• License session hosts with vApps per concurrent user, and with RTX vWS per session for professional 3D applications.
• Keep compute VMs on NVIDIA AI Enterprise per GPU, separate from the graphics editions.
• Choose the term against the hardware's life: a 4- or 5-year subscription or a perpetual licence with five-year SUMS covers a host bought now through most of its service.
• Check the vGPU branch: the RTX PRO 4500 Server Edition is supported only from vGPU 20, a branch supported until March 2027, while the RTX PRO 6000 Server Edition also runs on the long-term vGPU 19 branch, supported until July 2028, as our guide to RTX PRO 6000 Server Edition vGPU profiles shows.

Replacing older vGPU cards does not change the count either; our article on replacing A40 and A16 vGPU hosts maps old profiles onto the new cards.

Our software and licensing page lists the NVIDIA licences we supply, and our page on NVIDIA AI Enterprise and vGPU licences shows how each is counted. Describe your desktop pools and the term you prefer, and we put the licence lines next to the hardware in one quote.

FAQ