Migrating a VM configured with NVIDIA vGPU software release 20.1 to a host running release 20.0 fails#
Description#
If a VM configured with NVIDIA vGPU software release 20.1 is migrated to a host running NVIDIA vGPU software release 20.0, the migration fails and the VM shuts down.
This issue occurs only with the following combination of releases of guest VM graphics driver, vGPU manager on the source host, and vGPU manager on the destination host:
Guest VM Graphics Driver |
Source vGPU Manager |
Destination vGPU Manager |
|---|---|---|
20.1 |
20.1 |
20.0 |
Workaround#
Note
Tesla M10 GPUs do not support this workaround. Even after applying this workaround to a system on which this issue occurs, vGPU migration with Tesla M10 GPUs fails with the following error:
Unexpected migration data block encountered.
On the host that is running vGPU Manager 20.1, set the registry key
RMSetVGPUVersionMaxto 0x40001.Start the VM.
Confirm that the vGPU version in the log files is 0x40001.
2026-05-15T10:19:05.420Z| vthread-2142280| I125: vmiop_log: vGPU version: 0x40001
The VM can now be migrated.
Status#
Not a bug
Ref. ##
200533827