We are attempting to get vGPU-enabled guests working with our oVirt 4.3.0
configuration, but have run into problems.
We are running:
NVidia License Server version 2018.09
NVidia License Client Manager 2018.10.0.25098346
and that license info is correctly retrieved by the clients.
With a Centos 7.6 guest, we can ssh into the guest and see that the X
server is running and a gdm process is also running, but when we open a
remote console viewer, all we get is a black screen. The X server reports
the correct vGPU profile. The only error we see in the Xorg.0.log file is:
[ 5.176] (EE) NVIDIA(GPU-0): Failed to update the GLS device with FRL
configuration
but that is not fatal.
With a Windows 10 guest, we do get a login screen to appear, but the mouse
and keyboard do not seem to be functional.
Host configuration:
host is running oVirt-node 4.3.0
NVidia driver NVIDIA-vGPU-rhel-7.6-410.91.x86_64.rpm installed
CentOS 7.6 guest configuration:
NVidia driver NVIDIA-Linux-x86_64-410.92-grid.run installed
added an explicit BusID in the guest's xorg.conf file
disabled SELinux
Windows 10 guest configuration:
NVidia driver 412.16_grid_win10_server2016_64bit_international.exe
installed
Remote viewer:
remote-viewer 5.0-11.el7 running on CentOS 7.3
Both host and guest NVidia drivers come from the
NVIDIA-GRID-RHEL-7.6-410.92-410.91-412.16.zip package. We have tried with
both V100 and P4 GPUs (with ECC disabled), and encounter similar problems.
Any ideas on what to try, or how to better diagnose the cause of these
problems?
Thanks.
Show replies by date