NVIDIA vGPU hypervisor support by GPU: RTX PRO Server Edition, L40S, L4, H200 NVL (2026)

The matrix: which card runs vGPU on which hypervisor

NVIDIA supports vGPU for the RTX PRO 6000 Blackwell Server Edition, the RTX PRO 4500 Blackwell Server Edition, the L40S and the L4 on VMware vSphere, Windows Server 2025 with Hyper-V, Red Hat Enterprise Linux with KVM, Ubuntu, the Linux with KVM platforms (Proxmox VE, Nutanix AHV, Red Hat OpenStack Platform and others) and XenServer. The H200 NVL and the workstation editions of the RTX PRO 6000 are on none of these lists. This is what NVIDIA's validated-platforms pages for vGPU software R595, releases 20.0 to 20.2, state per hypervisor, each page last updated on 29 September 2026 and read by us on 25 October 2026.

The table shows what each page lists for the cards we supply. "Time-sliced" means vGPUs that share the whole card in turns; "MIG" means MIG-backed vGPU, where a vGPU sits on a hardware partition of the card, either alone (1:1) or time-sliced inside it.

NVIDIA Virtual GPU Software R595 validated platforms for VMware vSphere, Windows Server, Red Hat Enterprise Linux with KVM, Ubuntu, Linux with KVM and XenServer, each last updated 29 September 2026, read 25 October 2026. The L40S and L4 have no MIG. "Windows Server" covers Hyper-V on Windows Server 2025; Windows Server 2022 supports these cards by passthrough (DDA) only.

Linux with KVM has its own section below, because NVIDIA lists GPUs and features there but leaves version support to each hypervisor vendor. For the cards in the table, NVIDIA's Linux with KVM page shows the same entries as the RHEL and Ubuntu column: time-sliced and MIG-backed vGPU for both RTX PRO Server Editions, time-sliced vGPU for the L40S and L4.

This page is a hub. It answers which combination NVIDIA supports and links the guide for each platform. The setup steps, profiles and licence details are on those pages, which we name in each section.

Hypervisor versions NVIDIA lists for vGPU 20

A card being on a hypervisor's list is half the check; the hypervisor release has to be on it too. The versions below are those NVIDIA's R595 pages give, as of 29 September 2026.

NVIDIA vGPU R595 validated platforms per hypervisor and vGPU 20.0 to 20.2 release notes, updated 29 September 2026, read 25 October 2026. ESXi 9 is distributed only as part of VCF.

Two rows carry traps. On vSphere, ESXi 8.0 builds before Update 3 P06 are not supported at all, and the two RTX PRO Server Editions need at least 8.0 Update 3g, which our vSphere passthrough and vGPU guide explains with the build number. On Red Hat Enterprise Linux, NVIDIA lists only specific minor releases; release 20.2 added RHEL 10.2 and 9.8 and withdrew 9.7, so a host on 9.7 has to move before it can take 20.2. Which Linux to choose for a GPU host, including the kernel lines per release, is the subject of our article on Linux for GPU servers: Ubuntu, RHEL or Rocky.

MIG-backed vGPU per hypervisor

MIG splits the RTX PRO 6000 Server Edition into up to four instances of 24 GB and the RTX PRO 4500 Server Edition into two of 16 GB, as our guide to MIG and vGPU VMs per card sets out. Whether a VM can sit on such an instance depends on the hypervisor, and here the platforms differ more than for plain vGPU.

NVIDIA vGPU R595 validated platforms, columns "1:1 MIG-backed vGPU" and "Time-sliced, MIG-backed vGPU" for the RTX PRO 6000 and RTX PRO 4500 Blackwell Server Editions, updated 29 September 2026, read 25 October 2026.

On vSphere the VCF release decides. NVIDIA's footnote ties time-sliced MIG-backed vGPU, several vGPUs inside one instance, to VCF 9.1 and states that "all ESXi 8 releases are not supported" for that mode. For 1:1 MIG-backed vGPU, where one vGPU fills a whole instance, the page gives no release condition, but the 20.0 release notes list MIG-backed vGPUs for graphics among the features new on VCF 9.1; plan MIG on vSphere on VCF 9.1, and have NVIDIA confirm any other build first. Assigning several MIG-backed vGPUs to one VM came with release 20.2, again only on VCF 9.1. On Windows Server and XenServer NVIDIA marks both MIG columns for the Blackwell cards as not applicable, so a Hyper-V or XenServer host uses these cards with time-sliced profiles only. The density this costs is easy to put in numbers: with MIG and time-slicing combined the RTX PRO 6000 Server Edition holds up to 48 office desktops on 2 GB profiles per NVIDIA's vPC sizing guide, against 32 with time-sliced profiles alone, figures our RTX PRO 6000 Server Edition vGPU profiles guide sources.

We supply the RTX PRO 6000 Server Edition, the RTX PRO 4500 Server Edition, the L40S and the L4 as cards or in AI servers built to order; send us your hypervisor and release with the VM types you plan, and our team replies with the card and licence lines that match NVIDIA's lists.

Linux with KVM: Proxmox VE, Nutanix AHV, Red Hat OpenStack

NVIDIA files several products under "Linux with KVM". Its R595 page names H3C CAS, Inspur InCloud Sphere, Nutanix AHV, Proxmox VE, Red Hat OpenStack Platform, Sangfor aDesk, SUSE Linux Enterprise Server and Verge.io VergeOS, gives no version numbers and says: "For NVIDIA vGPU software support, contact your Linux with KVM hypervisor vendor." The GPU list on that page is the same set of current cards: both RTX PRO Server Editions with time-sliced and MIG-backed vGPU, the L40S and the L4 with time-sliced vGPU.

NVIDIA vGPU R595 Linux with KVM validated platforms and Nutanix AHV support page ("Nutanix AHV is supported on this release of NVIDIA vGPU software as a generic Linux with KVM hypervisor"), both updated 29 September 2026, read 25 October 2026. The last row is our reading: NVIDIA's lists name host operating systems, not upstream OpenStack.

For Proxmox VE the vendor's wiki carries the tested combinations of Proxmox release, kernel and vGPU driver, up to vGPU 20.2, and Proxmox calls its platform officially supported from vGPU 18; the subscription terms, the MIG steps and the limits are in our guide to NVIDIA vGPU on Proxmox VE. Nutanix distributes the vGPU Manager for AHV itself, through its own portal, as both NVIDIA's Linux with KVM page and the AI Enterprise 8.2 support matrix say; node choice and MIG questions on AHV are in our article on a Nutanix GPU cluster for AI. For OpenStack, our guide to NVIDIA vGPU in OpenStack with Nova and mdev covers the Nova configuration, flavors and the passthrough route for the H200 NVL.

Cards that are not on the vGPU lists

H200 NVL. None of the six R595 validated-platform pages covered here lists it. It is on the GPU list of NVIDIA AI Enterprise 8.2 (updated 2 September 2026), whose virtualised deployments run either in vGPU mode or in passthrough mode with the data-centre driver in the guest, on vSphere 8.0 and 9.0, Red Hat Enterprise Linux with KVM, Ubuntu and Nutanix AHV; the matrix does not map each GPU to a mode or hypervisor. For a VM plan we therefore treat the H200 NVL as a passthrough card, one or more whole cards per VM, and have any vGPU use confirmed with NVIDIA for the exact hypervisor release. With NVLink bridges there is one more rule from NVIDIA's RHEL KVM limitations: all GPUs joined by NVLink must go to the same VM, or the NVLink state is corrupted (XID 74) until the GPUs are reset.

RTX PRO 6000 Workstation and Max-Q editions. No vGPU list names them. They go into VMs by whole-card passthrough, and passthrough of these cards on multi-GPU hosts has reported reset problems, which our article on the RTX PRO 6000 Max-Q in virtual machines collects.

RTX PRO 5000 72 GB. Since release 20.2 it supports vWS and vPC on Red Hat Enterprise Linux with KVM 9.6, including MIG-backed vGPU, and nowhere else. The 48 GB card and the workstation RTX PRO 4500 and RTX PRO 4000 are not listed.

vGPU 19 or vGPU 20: which branch each card can use

NVIDIA's release table, read on 25 October 2026, shows two active branches: vGPU 20 on the R595 driver branch, a production branch with release 20.2 of August 2026 and end of life in March 2027, and vGPU 19 on R580, a long-term support branch with release 19.6 of August 2026 and end of life in July 2028. vGPU 18 and the R535-based vGPU 16 have reached end of life.

NVIDIA vGPU software release table and vGPU 20.0 to 20.2 release notes for vSphere and RHEL KVM, read 25 October 2026; first supported releases of the RTX PRO 6000 Server Edition, L40S and L4 as cited in our vSphere passthrough and vGPU guide.

The practical effect: a host with RTX PRO 4500 Server Editions stays on the shorter R595 branch and moves to NVIDIA's next branch within its support window. A fleet that wants the long-term branch to mid-2028 uses the air-cooled RTX PRO 6000 Server Edition, the L40S or the L4. Security fixes run inside each branch; vGPU 19.6 and 20.2 fix CVE-2026-47489 from NVIDIA's security bulletin of 30 September 2026, as our article on NVIDIA driver branches and CUDA versions records, so these two releases are the minimum.

Licences and compute profiles across hypervisors

The licence follows the profile series on every hypervisor: RTX vWS for Q-series, vPC or vWS for B-series, vApps for A-series, as our guide to NVIDIA vGPU licensing with vWS, vPC and vApps explains. Compute (C-series) profiles are outside NVIDIA's vGPU product. NVIDIA's RHEL KVM limitations state that "NVIDIA vGPU for Compute" is not available there and point to NVIDIA AI Enterprise, and the AI Enterprise 8.2 matrix lists vSphere, Red Hat Enterprise Linux with KVM, Ubuntu, Nutanix AHV and Red Hat OpenStack, with the KVM-based ones marked single-node only. Windows Server and XenServer are not on that list, and NVIDIA's Windows Server page lists only vWS, vPC and vApps for these cards.

For a mixed estate this gives a simple rule. Graphics desktops and virtual workstations run on any of the platforms above with the listed cards. Compute VMs on vGPU need AI Enterprise per GPU and one of its hypervisors; elsewhere they get whole cards by passthrough. Our AI Enterprise licensing guide has the per-GPU terms.

How to read NVIDIA's pages before you order

Each hypervisor has its own "Validated Platforms" page in NVIDIA's vGPU documentation, and each covers one driver branch, so the R580 pages for vGPU 19 and the R595 pages for vGPU 20 can differ for the same card. Check four things per combination: the card in the GPU table, the hypervisor release in the version list, the column for the vGPU mode you plan (single-instance time-sliced, 1:1 MIG-backed, time-sliced MIG-backed, passthrough), and the footnotes, where NVIDIA puts conditions such as VCF 9.1 for time-sliced MIG or RHEL 9.6 for the RTX PRO 5000. The pages also note that the RTX PRO 6000 Server Edition and the L40S ship in display-off mode and must be used in it for vGPU.

Older cards follow other dates. The T4 and A40 are in vGPU extended support until December 2027, and hosts built on them are a replacement question that our article on replacing A40 and A16 vGPU hosts handles. If your platform is not on the lists above, tell us the hypervisor, its release and the cards you have, and we check NVIDIA's current pages for that combination before you order.

What we supply

We supply the RTX PRO 6000 Server Edition, the RTX PRO 4500 Server Edition, the L40S, the L4 and the H200 NVL as cards with manufacturer warranty, or built into AI servers to order, assembled and tested before delivery. vGPU and NVIDIA AI Enterprise licences can go on the same quote as the hardware, as our licensing page describes; quantities and lead time are confirmed in writing with the quote, on one EU contract and invoice.

FAQ