Unable to determine the device handle for GPU 0000:00:09.0

hloewe · December 22, 2017, 1:29pm

I eventually succeed to install a ubuntu machine on k with device pass through

Cuda installed properly. After reboot nvidia driver a bound to the device

But when trying:

$ nvidia-smi
$ Unable to determine the device handle for GPU 0000:00:09.0

of course, in my case the card is not a X11 device. the boot vga adapter is the QEMU adapter.
But this is also the case when using the card on my local box without pass through.
Here the card is installed as the second card, laso without serving as a graphic device.

Any idea?

Regards

Heiko

Robert_Crovella · December 22, 2017, 1:34pm

What GPU are you passing through?

hloewe · December 22, 2017, 1:46pm

It is a GeForce GTX 1050 Ti

Robert_Crovella · December 22, 2017, 2:05pm

NVIDIA doesn’t support using GeForce GPUs in a pass-through virtualization mode, and the driver attempts to detect this condition.

So your experience is not surprising to me. It’s expected behavior that it doesn’t work.

hloewe · December 22, 2017, 2:12pm

that hurts :-(. But I can still use both cards fror Cuda on my local device. What is good :-).

BTW: which cards would be supported for pass thru, may be I buy one of those.

Thanks and Regards

Heiko

Robert_Crovella · December 22, 2017, 2:19pm

I don’t have a list that I can point you to. Most higher end Quadro and all Tesla adapters should work in a virtualized setting, generally speaking. Before making any purchase commitments, I would recommend you evaluate any GPU for desired behavior.