NVIDIA Tesla M60 not selectable for hardware transcoding in Docker

[eunux 0]

I know there are dozens of threads on this topic - I think I've read them all. I normally wouldn't post, but I spent an entire day trying to figure this out and I'm finally stuck. 

The Setup

• Dell R730 Poweredge server
• NVIDIA Tesla M60 "server" graphics card 
• Dual Intel Xeon E5 CPUs 
• Proxmox VE 8.1.4 running on the bare metal
• Ubuntu 22.04.4 LTS running as a VM
• Docker 25.0.3 running inside Ubuntu
• Emby (Premiere) 4.8.1.0

So obviously there are a lot of layers.  However, I successfully managed to get the NVIDIA M60 all the way through to Docker.  I can verify this by running the sample Docker workload provided by NVIDIA, from the Ubuntu host: 

docker run --rm --runtime=nvidia --gpus all ubuntu nvidia-smi

This works fine: 

This would suggest to me that, from the hardware all the way through to Docker, it's possible for the software running in the container to get a handle on the M60. This is confirmed in the emby logs that follow, where you can see e.g. ** "VendorName": "NVIDIA Corporation", "DeviceName": "GM204GL [Tesla M60]" ** in the logs, suggesting that the Emby code is getting pretty damn close to negotiating with the GPU.

I see the same driver version whether running this command via the nvidia docker image or just natively on my host terminal; this is probably expected but just confirming.

The Problem 

I've tried both the linuxserver.io and the official emby docker container, and neither of them give me the option to use the M60 for transcoding: 

Other Information

As you will see in both the hardware_detection and the embyserver logs, there is a lot of wrestling with the GPU, trying to set it up. But for whatever reason, it just fails. I've tried googling the various errors but haven't gotten anywhere. 

Container configuration

• The NVIDIA container runtime is installed 
• I've tried dozens of different configs in my docker-compose, but at this point I'm running with a simple config that seems to work for a lot of other people, and produces the same errors as all the more complicated setups I've tried. In this setup, group 44 is `video` and group 109 is `render` on the host. 

Any help is much appreciated. I am a few minutes away from giving up here

embyserver.txthardware_detection-63844435256.txt

Edited February 25 by eunux
fixing some formatting

[Luke 37071]

Hi, are you using the official nvidia drivers?

[eunux 0]

This works a little differently to how it used to, so I'm not totally confident I'll answer correctly BUT, I think so. 

I am using the drivers from the Nvidia-managed repository (not open source / canonical),  which I installed according to these instructions. 

[Luke 37071]

OK as a test, are you able to spin up our native package outside of Docker and see how that compares? That would help narrow things down.

[oliveiramarinho 0]

The problem was solved? I'm having the same problem and I'm using the same version.

[Luke 37071]

On 3/21/2024 at 10:30 AM, oliveiramarinho said:

The problem was solved? I'm having the same problem and I'm using the same version.

Hi @oliveiramarinhocan you please describe the issue that you're having? What have you done so far?