Isnt proxmox debian though? So would I install apt install nvidia-driver? Because apt install nvidia-driver-550 on the host can’t be found.
OK, following this guide, Nvidia GPU passthrough in LXC :: TheOrangeOne I was able to install the official NVIDIA drivers and after rebooting, nvidia-smi works.lxc command for your script to run though. Update: Looks like the new command is pct
Update 2: Continuing to follow that guide, I actually have nvidia-smi working in the LXC. I had to remove nomodeset from the grub command line, blacklist the nouveau drivers in proxmox, and run nvidia-persistenced to create the /dev/nvidia-modeset character device. Next step is to connect my truenas NFS share and try plex.
Chuck, I was able to get Plex on the LXC but it doesn’t show the GPU in the list, yet nvidia-smi shows results. What steps should I try next?
EDIT: I finally got it!
ctid=100 # change to plex container id
renderid=$(pct exec $ctid getent group render | cut -d: -f3)
pct set $ctid -dev0 /dev/dri/renderD128,gid=$renderid
pct reboot $ctid
That got the gpu to show up in the list.
Then I had to add the plex user to the lxc_shares group (to show my media. All from this guide: [TUTORIAL] - Tutorial: Unprivileged LXCs - Mount CIFS shares | Proxmox Support Forum
And it worked. Odd thing is, yes it shows (hw) in plex dashboard and my CPU is doing almost nothing, but nvidia-smi doesn’t show anything:
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.142 Driver Version: 550.142 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Quadro P400 Off | 00000000:81:00.0 Off | N/A |
| 34% 46C P0 N/A / N/A | 140MiB / 2048MiB | 1% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
+-----------------------------------------------------------------------------------------+
ChuckPa
January 11, 2025, 5:22pm
24
I used the ‘tteck’ helper scripts to create mine.
When you look at the config, this is what you see:
root@pve:~# pct config 100
arch: amd64
cores: 4
features: nesting=1
hostname: plex
memory: 4096
mp0: /glock,mp=/glock
net0: name=eth0,bridge=vmbr0,hwaddr=BC:24:11:64:46:69,ip=dhcp,type=veth
onboot: 1
ostype: ubuntu
rootfs: ContainersVMs:vm-100-disk-0,size=256G
swap: 512
tags: proxmox-helper-scripts
lxc.cgroup2.devices.allow: a
lxc.cap.drop:
lxc.cgroup2.devices.allow: c 188:* rwm
lxc.cgroup2.devices.allow: c 189:* rwm
lxc.mount.entry: /dev/serial/by-id dev/serial/by-id none bind,optional,create=dir
lxc.mount.entry: /dev/ttyUSB0 dev/ttyUSB0 none bind,optional,create=file
lxc.mount.entry: /dev/ttyUSB1 dev/ttyUSB1 none bind,optional,create=file
lxc.mount.entry: /dev/ttyACM0 dev/ttyACM0 none bind,optional,create=file
lxc.mount.entry: /dev/ttyACM1 dev/ttyACM1 none bind,optional,create=file
lxc.cgroup2.devices.allow: c 226:0 rwm
lxc.cgroup2.devices.allow: c 226:128 rwm
lxc.cgroup2.devices.allow: c 29:0 rwm
lxc.mount.entry: /dev/fb0 dev/fb0 none bind,optional,create=file
lxc.mount.entry: /dev/dri dev/dri none bind,optional,create=dir
lxc.mount.entry: /dev/dri/renderD128 dev/dri/renderD128 none bind,optional,create=file
root@pve:~#
You’ll need to add the nvidia entries as well . ( I’ve seen them posted here in the forum but not sure where )
I use these three. You might need a fourth (IIRC)
nvidia.driver.capabilities: all
nvidia.require.cuda: "true"
nvidia.runtime: "true"
I’ll have to try this. When I am hardware transcoding, and open nvidia-smi on the host, this is what I see:
root@pve:/etc/pve/lxc# nvidia-smi
Sat Jan 11 13:14:01 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.142 Driver Version: 550.142 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Quadro P400 On | 00000000:81:00.0 Off | N/A |
| 34% 39C P0 N/A / N/A | 140MiB / 2048MiB | 33% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 346273 C ...lib/plexmediaserver/Plex Transcoder 135MiB |
+-----------------------------------------------------------------------------------------+
Which definitely looks correct.
ChuckPa
January 11, 2025, 6:32pm
26
That is absolutely correct.
So what’s not working?
Can you capture me a set of DEBUG logs (ZIP) right after it starts transcoded playback?
I tried your lxc config above and hw transcoding did not work, it fell back to software.
# Allow cgroup access
# Pass through device files
arch: amd64
cores: 4
dev0: /dev/dri/renderD128,gid=993
features: nesting=1
hostname: plex
memory: 8192
mp0: /media/truenas/videos/,mp=/media,ro=1
mp1: /media/truenas/photos/,mp=/photos,ro=1
net0: name=eth0,bridge=vmbr0,firewall=1,hwaddr=BC:24:11:C7:9D:12,ip=dhcp,ip6=dhcp,type=veth
ostype: ubuntu
rootfs: local-lvm:vm-102-disk-0,size=80G
swap: 1024
unprivileged: 1
lxc.cgroup2.devices.allow: c 195:* rwm
lxc.cgroup2.devices.allow: c 509:* rwm
lxc.mount.entry: /dev/dri/card0 dev/dri/card0 none bind,optional,create=file
lxc.mount.entry: /dev/nvidia0 dev/nvidia0 none bind,optional,create=file
lxc.mount.entry: /dev/nvidiactl dev/nvidiactl none bind,optional,create=file
lxc.mount.entry: /dev/nvidia-uvm dev/nvidia-uvm none bind,optional,create=file
lxc.mount.entry: /dev/nvidia-modeset dev/nvidia-modeset none bind,optional,create=file
lxc.mount.entry: /dev/nvidia-uvm-tools dev/nvidia-uvm-tools none bind,optional,create=file
lxc.mount.entry: /dev/nvram dev/nvram none bind,optional,create=file
Doing nvidia-smi on the host when it’s transcoding works. Doing it in the lxc, while shows an output, does not show any processes.
vm 102 - unable to parse config: nvidia.driver.capabilities: all
vm 102 - unable to parse config: nvidia.require.cuda: "true"
vm 102 - unable to parse config: nvidia.runtime: "true"
ChuckPa
January 11, 2025, 7:16pm
28
I use LXD / LXC,
The config looks like this:
[chuck@lizum ~.2000]$ lxc config show nvidia
architecture: x86_64
config:
image.architecture: amd64
image.description: ubuntu 22.04 LTS amd64 (release) (20240821)
image.label: release
image.os: ubuntu
image.release: jammy
image.serial: "20240821"
image.type: squashfs
image.version: "22.04"
nvidia.driver.capabilities: all
nvidia.require.cuda: "true"
nvidia.runtime: "true"
volatile.base_image: a3a8118143289e285ec44b489fb1a0811da75c27a22004f7cd34db70a60a0af4
volatile.cloud-init.instance-id: cbebf9f6-289d-4fde-af56-0bba1dfecd12
volatile.eth0.hwaddr: 00:16:3e:15:d5:5a
volatile.idmap.base: "0"
volatile.idmap.current: '[{"Isuid":true,"Isgid":false,"Hostid":1000000,"Nsid":0,"Maprange":1000000000},{"Isuid":false,"Isgid":true,"Hostid":1000000,"Nsid":0,"Maprange":1000000000}]'
volatile.idmap.next: '[{"Isuid":true,"Isgid":false,"Hostid":1000000,"Nsid":0,"Maprange":1000000000},{"Isuid":false,"Isgid":true,"Hostid":1000000,"Nsid":0,"Maprange":1000000000}]'
volatile.last_state.idmap: '[]'
volatile.last_state.power: STOPPED
volatile.uuid: 11e2f94c-1046-473c-b219-ae051f8aa797
volatile.uuid.generation: 11e2f94c-1046-473c-b219-ae051f8aa797
devices:
gpu:
gid: "110"
type: gpu
media:
path: /glock/media
readonly: "true"
source: /glock/media
type: disk
ephemeral: false
profiles:
- default
stateful: false
description: ""
[chuck@lizum ~.2001]$
Are you using TrueNAS Core (BSD) or are you using TrueNAS Scale (Linux) ?
What is your host OS? I don’t have the lxc command.
I am using TrueNAS Core (BSD).
ChuckPa
January 11, 2025, 7:46pm
30
I’m sorry, everything I’ve been sharing is for naught.
For you, you have Jails ?
Yes, I’m moving plex from a truenas jail to proxmox VM on a different server. I drastically expanded my homelab solution over the past few weeks to add to my proxmox and didnt want to interfere with messing with my truenas.
ChuckPa
January 12, 2025, 6:56pm
32
Setting up a new machine from scratch is the best bet. There’s little/nothing in common between the two except, perhaps, source media locations.
As for Prox and HW transcoding, while I don’t have a GPU card, I do have the iGPU support via those helper scripts which perform the passthrough commands
Very sorry but this is the best I can do is “Mr Google”…
https://www.reddit.com/r/Proxmox/comments/1cneob0/plex_lxc_gpu_passthrough/
system
April 12, 2025, 6:56pm
33
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.
