CVE-2024-0084 is a High severity (CVSS 7.8) flaw in the NVIDIA Virtual GPU Manager on Linux, where the guest operating system could execute privileged operations. NVIDIA fixed it in vGPU 17.2, 16.6 and 13.11 for the hypervisors it lists. Update the Virtual GPU Manager on every host to the release for your branch.
What's affected
- Product: NVIDIA vGPU software, Virtual GPU Manager (host side), with the CPE entries pointing to Ubuntu hosts. The bulletin also lists Citrix Hypervisor, VMware vSphere and Red Hat Enterprise Linux KVM.
- Fixed Virtual GPU Manager releases: 17.2 (550.90.05), 16.6 (535.183.04) and 13.11 (470.256.02). Cloud Gaming: May 2024 release, 555.52.04 for the Virtual GPU Manager on Red Hat Enterprise Linux KVM and VMware vSphere.
- Severity: High, CVSS 7.8, local vector, low privileges. NVIDIA lists impacts of information disclosure, data tampering, privilege escalation and denial of service. The record and the bulletin agree.
- NVIDIA notes that its tables may not be exhaustive. If you run an earlier branch with no listed fix, move to the latest branch release.
- Exploitation: the record's KEV flag is false, and the bulletin does not report exploitation.
How to check if you're exposed
On each vGPU host, run:
nvidia-smi
and read the driver version at the top. Compare it with the fixed Virtual GPU Manager versions above for your branch. Also list which hypervisor and which vGPU release branch (13, 16 or 17) each host uses, since the fixed version depends on both.
Guest virtual machines are the attack source in this bug, so note which tenants or teams have guests on each host.
What to do now
- List vGPU hosts, their Virtual GPU Manager version and release branch.
- Download the fixed release from the NVIDIA Licensing Portal.
- Plan a host maintenance window, since updating the manager usually means migrating or stopping guests.
- Update guest drivers to the matching release at the same time (see the related driver advisory).
- Restrict who can create guest VMs with vGPU profiles on shared hosts.
Why it matters
The record maps this to CWE-250 (CWE-250), execution with unnecessary privileges. The Virtual GPU Manager sits on the host and serves guests. If a guest can ask it to do something privileged, the boundary between virtual machines and the host is weaker than the design intends.
This matters most for shared infrastructure, such as a virtual desktop farm or a GPU cloud, where the guest owner is not the host owner. Single tenant labs have less exposure. Either way, the bulletin gives clear fixed versions, so the work is scheduling, not research.
Quick check
Which vGPU releases fix CVE-2024-0084?
17.2, 16.6 and 13.11, with Virtual GPU Manager versions 550.90.05, 535.183.04 and 470.256.02 for the listed hypervisors.
Is the guest or the host the part to patch?
For this CVE the bulletin lists the Virtual GPU Manager on the host. Update guest drivers too for the related CVEs in the same bulletin.
Sources
- NVIDIA security bulletin 5551 (vGPU and GPU display driver, June 2024)
- Safeguard Gold entry for CVE-2024-0084
Last reviewed: 19 October 2026.
