Hi all,
I’m testing NVIDIA GH100 (H200 NVL) GPUs on Proxmox VE 9.0.3 with the vGPU driver package:
NVIDIA-Linux-x86_64-570.195.02-vgpu-kvm.run
Current Situation:
After installation, I’m experiencing the following issues:
nvidia-vgpu-mgrservice fails to start with the error:Failed to attach device: 0x59 (gpuId 0xae00)/sys/class/mdev_bus/and/sys/bus/pci/devices/.../mdev_supported_types/directories are missingnvidia-smi -qshows:Virtualization Mode : Host VGPU Host VGPU Mode : SR-IOV sriov_totalvfs : 32
Observations:
Based on these results, it appears that the H200 only supports SR-IOV-based vGPU virtualization and not the traditional mediated device (mdev) mode. When testing under Proxmox, the pve-nvidia-sriov@ services work correctly and VF devices are created successfully, but no mdev profiles are available, and nvidia-vgpu-mgr consistently fails to attach the device.
Question:
Can anyone confirm whether mdev-based vGPU is deprecated or unsupported on GH100/H200 series GPUs? Is there any official documentation or compatibility matrix that clarifies which GPU models still support mdev versus SR-IOV-only configurations?
Any guidance or links to official NVIDIA documentation would be greatly appreciated.
Thanks in advance!