GPU alloctaion - Truecharts Plex app

[stealthbob]

I recently started getting transcoding issues from my family on my Plex server, unfortunately due to them not wanting to complain I am unsure when it happened. I can only assume that it started when I updated to Cobia.

Thing is I have a 1050Ti installed but was never able to get it actually working for transcoding.

Yes I have Plex pass

Seems the system recognizes the GPU

I have isolated the GPU in the settings

I dont se any options to pick for GPU passthrough:

That means I dont see any devices in the Plex transcoding settings as well (dropdown only says Auto)

I have never required transcoding but recently users are buffering and at times getting the "This server is not strong enough" type plex nag.

I have been over this for quite some time...hoping that I missed some realy simple thing.

Thanks for any comments you may have.

 

[joace]

if need to assign GPU to apps, supposedly you don't need to isolate it in config, that's for VM assignment i guess.
But even with correct setting, I still have the same issue not getting GPU working in apps, no process detected, previously they said it's an issue with iommu in 23.10.0.1, but after upgrading to the latest, still the same.

 

[joace]

After reading some forum threads, I'm guessing this is because the Nvidia driver picked up in Cobia is actually data center version, which has much less supported hw, especially consumer lines, e.g., below are supported hw by this driver:

L-Series:
L40, L40S, L4

H-Series:
NVIDIA H100 PCIe, NVIDIA H100 NVL, NVIDIA H800 PCIe, NVIDIA H800 NVL

A-Series:
NVIDIA AX800, NVIDIA A800, NVIDIA A100, NVIDIA A40, NVIDIA A30, NVIDIA A16, NVIDIA A10, NVIDIA A2

RTX-Series:
RTX 8000, RTX 6000, NVIDIA RTX A6000, NVIDIA RTX A5000, NVIDIA RTX A4000, NVIDIA T1000, NVIDIA T600, NVIDIA T400

HGX-Series:
HGX H800, HGX H100, HGX A800, HGX A100, HGX-2

T-Series:
Tesla T4

V-Series:
Tesla V100

P-Series:
Tesla P100, Tesla P40, Tesla P6, Tesla P4

M-Class:
M60, M40 24GB, M40, M6, M4

 

[stealthbob]

Thanks, that is great info....

More and more I keep telling myself I just need to stop using TNS as a hypervisor. The dependancy on others to not make system breaking changes has been frustrating...looking in your direction Truecharts. I appreciate all this is freeware and I rspect those who work hard at this ecosysem but I am going to leave TNS alone for what it does best...just a NAS. I am certainly not the end user type devs are thinking baout which is fair.

Honestly tired of this, I am moving everything off to bare metal...this time with a good Quicksync iGPU setup. After looking into it I am impressed with the abilities with current gen Intel in this regard. I can still employ my 1050Ti in that box as well with Tdarr as its fairly straight forward to expose GPUs this way.

Hope this issue is resolved for all those waiting, I will be moving along from using apps in TNS

 

[HoneyBadger]

After reading some forum threads, I'm guessing this is because the Nvidia driver picked up in Cobia is actually data center version, which has much less supported hw, especially consumer lines, e.g., below are supported hw by this driver:

We aren't using the datacenter driver, just the standard Linux x64 535.54.03 release; still supports consumer cards back to Maxwell.

Driver Details | NVIDIA

Download the <dd~LanguageName> <dd~Name> for <dd~OSName> systems. Released <dd~ReleaseDateTime>

www.nvidia.com

 

[joace]

We aren't using the datacenter driver, just the standard Linux x64 535.54.03 release; still supports consumer cards back to Maxwell.

Driver Details | NVIDIA

Download the <dd~LanguageName> <dd~Name> for <dd~OSName> systems. Released <dd~ReleaseDateTime>

www.nvidia.com

Good to hear, then there must be other issue underlying not sorted out, i had my 1060 worked in an old release, couldn't recall the exact version, but should before cobia, after upgraded, it just stopped working. Is there any tricks to triage where is problem?

 

[FrostyCat]

Here's a few more things to try.

Disable the GPU isolation as it's no needed.
Check if you have ticked "Enable GPU support" in the "Apps" / "Advanced Settings"

Kubernetes doesn't have native support for GPUs, it relies on device plugins to enumerate and lable nodes with GPUs available, so if this device plugin is not running, Kubernetes will not see your GPU, even if you have the correct drivers installed and enabled. The setting above should install this device plugin for you.

I don't have an Nvidia GPU myself, but, you could also get a shell on the machine and do a "k3s kubectl get pods -A | grep nvidia" and see if there's anything running.

 

[joace]

Thanks for your tips!

I can confirm that I don't isolate GPU and it's also enabled in app's k8s settings. The plugin is running from cli checking, since my nvidia 1060 has been recognized by the driver (shown in nvidia-smi) and I'm also able to assgin GPU to each app, but no any process shown to use the GPU, that's why I'm totally lost where to look at the issue.

root@truenas[~]# k3s kubectl get pods -A | grep nvidia
kube-system nvidia-device-plugin-daemonset-857x2 1/1 Running 0 3d20h

 

[FrostyCat]

In this case the node should have been labeled by the device plugin, can you try and run a "k3s kubectl describe node ix-truenas | grep gpu" and check the "Labels", "Capacity" and the "Allocatable" section.

Also, at the end of the "describe" output, check the "Allocated resources" section, if the GPU is already allocated it may not be shared to another app again.

Here's how it looks for me with an AMD GPU.