[ovirt-users] Nvidia Tesla M60 Direct GPU-Passthrough giving issues + VNC

[Martin Polednik]

On 30/01/18 11:15 +0800, Ravyu Sivakumaran wrote:
>Hi,
>
>I am running an Ovirt host on a SuperMicro SuperServer 2028GR-TRH equipped
>with 2x2 Tesla M60s (they are dual GPU Cards). I've followed this guide-
>https://www.ovirt.org/develop/release-management/features/virt/hostdev-passthrough/
>and managed to get a Windows 10 Enterprise guest running with one GPU
>attached to it. The windows guest successfully detects the single M60 GPU
>and I can install the latest Tesla drivers (390.65) on it with no problem.
>However, the GPU refuses to do any work whatsoever. The control panel says
>the GPU is being used for compute and not as a VGA adapter (shows up as
>inactive on the activity monitor). Nvidia's GPUModeSwitcher tool tells me
>that the GPU is in Graphics mode, so there is a driver issue here. I have
>access to the GRID Drivers. However I'm not sure why I'd need them as I'm
>only doing a direct-passthrough, not GPU Virtualization. Any ideas?

This might be an interesting question for NVIDIA - when using GPU
assignment for graphics, you'd expect the GPU to show grapics output
to a screen directly connected to the machine. Since M60 doesn't even
have any physical output, the drivers for it may not support
`graphical` mode outside of GRID setup.

What is your goal for the Tesla assignment?

>Also, after activating Device-Passthrough, I'm unable to access my VM's via
>the Hosts' VNC anymore. I get an "unable to connect" error. Im currently
>accessing the Guests with a direct VNC connection. Problem is, I can't
>clone/create new VMs anymore because of this. I must've done something
>wrong, not sure what.

Does the same happen in case of SPICE?

>Thanks in advance!
>-- 
>Yours Faithfully
>Ravyu

>_______________________________________________
>Users mailing list
>Users at ovirt.org
>http://lists.ovirt.org/mailman/listinfo/users