Emby randomly stops using Nvidia GPU and switches to on-board Intel

[txrxsystems 1]

Hey all, I'm a long-time lurker, first time poster on the forum.

I've got a Dell Precision Rack R3930 running Ubuntu server 20.04 (bare metal) with an Arr stack, NZBGet and Emby running on it. 

I have an Nvidia Tesla P4 GPU installed in it and it shows up just fine in the system and Emby detects it too. I have selected the Tesla as the primary transcoding GPU by dragging the Nvidia entries to the top of each list.

The problem is that randomly I'll notice that the machine is running it's fans very loud so I ssh into the box and try running htop and nvtop to see how hard the CPU/GPU are working. Sometimes nvtop will return the error "no gpu to monitor"

If I go into the Emby dashboard, under transcoding I will see that the Tesla entries are missing.

So far the only way I can find to fix it is to reinstall the Nvidia drivers/modules and reboot. This is happening almost every week so I'm getting more than a little tired of it.

Any ideas what I can do to stop this?  

[GrimReaper 3311]

1 hour ago, txrxsystems said:

Sometimes nvtop will return the error "no gpu to monitor"

If I go into the Emby dashboard, under transcoding I will see that the Tesla entries are missing.

That sounds more like system issue than Emby issue but nevertheless, post hardware detection log and ffmpeg log(s) when you observe reoccurrence. 

[txrxsystems 1]

OK I'll do that. 

I've just fixed it so I will monitor it closely and post the logs when it re-occurs.

Thanks

[txrxsystems 1]

Hey @grimreaper0744The problem has re-occurred already. 

I've grabbed a copy of the most recent Emby ffmpeg log and the Emby log but where do I find the hardware detection log?

[txrxsystems 1]

The ffmpeg log is attached. It contains an error relating to CUDA (strange that it's trying to use Cuvid when I have Nvenc as the first option in all my lists under Transcoding?)

Any advice on how I should go about finding the root cause?

ffmpeg-transcode-750e48fc-1cd9-48b1-bafb-db75f7d8a51b_1.txt

Edited May 22, 2023 by txrxsystems

[GrimReaper 3311]

On 5/22/2023 at 5:03 PM, txrxsystems said:

strange that it's trying to use Cuvid when I have Nvenc as the first option in all my lists under Transcoding?

It ain't using CUVID, NVENC/NVDEC are used, CUDA is just denoting using CUDA cores/api. 

Quote

>>>>>>  FindVideoEncoder - MediaType: h264, UseHardwareCodecs: True, HWA-Mode: Advanced
Info    Checking: 'NVENC Tesla P4 - H.264 (AVC)'
Info    Check successful - selecting 'NVENC Tesla P4 - H.264 (AVC)'

>>>>>>  FindVideoDecoder - MediaType: h264, UseHardwareCodecs: True, HWA-Mode: Advanced
Info    Checking: 'NVDEC Tesla P4 - H.264 (AVC)'
Info    Check successful - selecting 'NVDEC Tesla P4 - H.264 (AVC)'

 

On 5/22/2023 at 5:03 PM, txrxsystems said:

Any advice on how I should go about finding the root cause?

Exactly what you thought as odd: try putting CUVID as top preferred HW decoder and compare results, as there is no one-fits-all solution: on some systems NVDEC works better, on others CUVID does, that's why there are options. 

[softworkz 3341]

@txrxsystems - The Nvidia drivers provided by Ubuntu are known to be unstable. Please try to switch to the proprietary drivers (also provided by Ubuntu) or install the drivers from the Nvidia website, running their installer.

[Luke 37119]

@txrxsystemshas this helped?