[SOLVED] Nvidia LXC Setup

[TheDarthFather]

Hello,
After I installed the Nvidia drivers on proxmox I continued to use this to install it on the container itself.

First observation is that in my case (vs the refferenced guide ^^), on the HOST I am missing the dev/nvidia-modeset

root@hypervisor:~# ls -l /dev/nvidia*
crw-rw-rw- 1 root root 195,   0 Nov 13 14:10 /dev/nvidia0
crw-rw-rw- 1 root root 195, 255 Nov 13 14:10 /dev/nvidiactl
crw-rw-rw- 1 root root 510,   0 Nov 13 14:10 /dev/nvidia-uvm
crw-rw-rw- 1 root root 510,   1 Nov 13 14:10 /dev/nvidia-uvm-tools


/dev/nvidia-caps:
total 0
cr-------- 1 root root 238, 1 Nov 13 14:10 nvidia-cap1
cr--r--r-- 1 root root 238, 2 Nov 13 14:10 nvidia-cap2

I decided to skip over the problem for now .... and I installed the driver in the container.
While installing I got this as a warning:


After install:

root@plex:/opt/temp# nvidia-smi
Failed to initialize NVML: Unknown Error

I would appreciate any hints/guidance ...

Thank you!

 

[jasonsansone]

Hello,
After I installed the Nvidia drivers on proxmox I continued to use this to install it on the container itself.

First observation is that in my case (vs the refferenced guide ^^), on the HOST I am missing the dev/nvidia-modeset

root@hypervisor:~# ls -l /dev/nvidia*
crw-rw-rw- 1 root root 195,   0 Nov 13 14:10 /dev/nvidia0
crw-rw-rw- 1 root root 195, 255 Nov 13 14:10 /dev/nvidiactl
crw-rw-rw- 1 root root 510,   0 Nov 13 14:10 /dev/nvidia-uvm
crw-rw-rw- 1 root root 510,   1 Nov 13 14:10 /dev/nvidia-uvm-tools


/dev/nvidia-caps:
total 0
cr-------- 1 root root 238, 1 Nov 13 14:10 nvidia-cap1
cr--r--r-- 1 root root 238, 2 Nov 13 14:10 nvidia-cap2

I decided to skip over the problem for now .... and I installed the driver in the container.
While installing I got this as a warning:
View attachment 31338

After install:

root@plex:/opt/temp# nvidia-smi
Failed to initialize NVML: Unknown Error

I would appreciate any hints/guidance ...

Thank you!

What is your config for the LXC container? Almost all guides online are out of date - they do not provide for the change to v2 groups.

 

[TheDarthFather]

What is your config for the LXC container? Almost all guides online are out of date - they do not provide for the change to v2 groups.

 

[jasonsansone]

Stop the container. Replace all instances of:
lxc.cgroup.devices.allow

with:
lxc.cgroup2.devices.allow

Start container. What is the output now of ls -al /dev/nvidia* from within the container?

 

[TheDarthFather]

Wow .. instant success! Thank you!

Quick questions:
1. How do I realy know is working ? I runned a movie on plex now and it seems the CPU is under 5% so seems ok but it deos not seem to transcode ... but is there any specific test to make ?
Later Edit: Found the way to force a transcode - looks like is working!!!!
2. After the install of driver on host I lost my GPU temperature sensor in glances ... any way to "get it back" ?

 

[jasonsansone]

Wow .. instant success! Thank you!

Quick questions:
1. How do I realy know is working ? I runned a movie on plex now and it seems the CPU is under 5% so seems ok but it deos not seem to transcode ... but is there any specific test to make ?
2. After the install of driver on host I lost my GPU temperature sensor in glances ... any way to "get it back" ?

View attachment 31345

1. You could use handbrake and do an encode using nvenc. The encode would fail if nvenc isn’t available. You can also run the nvidia-smi command on the host to see all running processes. Lastly, Plex dashboard clearly indicates when it is using hardware vs software transcoding.

2. Not sure. That all feels very motherboard and GPU dependent. I have gpu temp monitoring through IPMI without issue but I have Tesla cards in server class motherboards.

 

[TheDarthFather]

1. You could use handbrake and do an encode using nvenc. The encode would fail if nvenc isn’t available. You can also run the nvidia-smi command on the host to see all running processes. Lastly, Plex dashboard clearly indicates when it is using hardware vs software transcoding.

2. Not sure. That all feels very motherboard and GPU dependent. I have gpu temp monitoring through IPMI without issue but I have Tesla cards in server class motherboards.

1. Indeed I forced a transcode and I see the hw, so it looks good!

Thank you again!

 

[ThijmenGThN]

1. Indeed I forced a transcode and I see the hw, so it looks good!

Thank you again!

Hi @TheDarthFather,

Quick question, did you manage to get "dev/nvidia-modeset" to show up again?

Mine went missing after accidentally mounting the GPU onto a VM (normally I pass it through LXCs), I've tried numerous solutions but haven't gotten it to work unfortunately.

^ Thijmen