Nvidia grid K1 multigpu usage with X and opengl


We are testing the new Cisco C240 rack server equipped with two Nvidia grid K1 cards (however we will only use one card later).

I’ve got X up and running, however it seems like I only use one gpu (looking at nvidia-smi, I will only see utilization on one gpu).

I thought that with the option, Option “MultiGPU” “Auto” in my xorg.conf, that the nvidia driver would make it possible to utilize all the available gpu’s ? Starting X with that option throws the following error though, which makes me think that something isn’t right,

(EE) NVIDIA(0): Failed to find a valid Multi-GPU configuration.
(WW) NVIDIA(0): Failed to find a valid Multi-GPU configuration for the NVIDIA
(WW) NVIDIA(0): graphics device PCI:9:0:0. Please see Chapter 28:
(WW) NVIDIA(0): Configuring SLI and Multi-GPU FrameRendering in the README
(WW) NVIDIA(0): for troubleshooting suggestions.
(EE) NVIDIA(0): Only one GPU will be used for this X screen.

I’m able to utilize all gpu-power if I’m using a xorg.conf that has all gpu’s defined as separate x screens, and then when invoking my command (eg. heaven benchmark / glxgears / etc…) I specify the display to run on. Then I will see utilization on the gpu’s that I’ve actually started the benchmarks/commands on. And I guess that’s fine, I just thought that I wouldn’t need to manually specify which gpu’s to use.

Am I missing something ?
I’ll happily answer any questions (if I’m not making my self clear), take any suggestions or observations.

Best regards,
Patrik Martinsson,

nvidia-bug-report.log (926 KB)

deleted …


Using separate X screens is the right way to use multiple GPUs in this case for GRID gpus.

Patchon, I’m trying to implement the grid with CentOS for an experiment. We just put a couple K1s in a GRID server today. I started work trying to get X to start using a passed-through GPU (KVM-QEMU). However, all we see as part of the X11 vnc is a black screen with nothing rendered. It seems like it may be my xorg, but I’m not sure if I completely understand how the K1 works. My understanding was that you could use the passed through GPU as a display processor and use a vncviewer to view the output from the X11 vnc server.