NVIDIA AI Enterprise
and vGPU licences
The licences are not counted the way most people assume: per GPU in one case, per concurrently active user in the other. Getting that wrong is expensive in both directions.
These are the licences that decide whether the card you bought can do the job you bought it for. We put them in the quote with the hardware, not after it.
| Licence | Counted by | Needed for | Note |
|---|---|---|---|
| NVIDIA AI Enterprise | Per GPU | The supported AI software stack, on DGX and RTX PRO Server | Bundled with some systems, separate with others. |
| RTX vWS (Virtual Workstation) | Per concurrently active user | Full workstation graphics inside a VM | Not per card: this is where budgets get misread. |
| Virtual PC / Virtual Apps | Per concurrently active user | Desktop and application streaming | vApps streams one display at 1280×1024. |
| MIG | Not licensed | Hardware partitioning of the card | Free. But with MIG enabled, NVLink is off and NCCL is unsupported. |
All licences above are available to order. Lead time is confirmed with every quote.
One counterparty in Vienna. No cross-border purchase to arrange, no import to clear on your side.
Where a card needs a licence to do the job you bought it for, it is in the quote, not discovered three months later.
Our partner Vixen.UNO designs, deploys and supports it. Under the same contract, or not at all – your call.
Manufacturer warranty handled through us, with support alongside the team who runs the system.
Then it drops to 1280×1024 at three frames per second with CUDA off. After reconnection the limits lift, but applications need restarting. This is documented behaviour, not a fault, and it is the single most common surprise we get called about.
Read the full analysis →MIG cuts the card in hardware: each instance gets its own compute blocks and its own memory bandwidth. vGPU slices time and needs a licence per user. RTX PRO 6000 Blackwell splits into four 24 GB instances; the L40S has no hardware partitioning at all.
Read the full analysis →NVIDIA quotes two to four heavy users per card, three to six medium, six to eight light. The sizing guide also notes the best-effort scheduler often allows two- to threefold oversubscription when users are not all active at once.
Read the full analysis →Is AI Enterprise per GPU or per server?
Do I need a licence for MIG?
What happens when a licence expires?
Which hypervisors support MIG-backed vGPU?
Can I buy licences and hardware on one contract?
The real per-card numbers, the licensing that counts users not cards, and where the setup breaks.
Read the article →Which nodes take cards, what works on AHV, and the limits that break half of all multi-GPU plans.
Read the article →Only one of the three got vGPU: the full comparison of what each edition can do.
Read the article →We reply within one business day