#CPU usage high despite GPU transcoding

1 messages · Page 1 of 1 (latest)

analog pier
#

Tunarr version 1.2.9
Running in Docker on unRAID

I've passed my GPU through to Tunarr which should be confirmed by image two showing Tunarr as an active app on the GPU, however when streaming a channel my CPU usage stays high consistently.

For comparison I've included screenshots running ErsatzTV. GPU was passed through to ErsatzTV the same way and seemingly successfully. When starting a stream CPU usage rises and falls after a short while; Tunarr keeps CPU usage high seemingly no matter how long the stream is open.

In this example both programs are transcoding the same media and the FFMPEG settings are the same other than Ersatz using NVENC and Tunarr using CUDA.

Am I missing a further setting or is this just a truth of how Tunarr runs on the CPU?

clever topaz
#

We would need a bit more information to help out:

  1. Transcoding configuration for the channel
  2. Full server logs when you’re playing the item
  3. How you’ve set Tunarr up to use your GPU (e.g. Docker compose file or similar)
analog pier
#

apologies, I didn't think logs would be useful for this issue.

Attached a screenshot of transcode config

GPU was added through the unRAID configuration for the app. I'm not sure if this counts as a docker compose since it's done in the unRAID GUI exclusively with the app running in a docker container. I added "--runtime=nvidia" to "Extra parameters" and then added the variables NVIDIA_VISIBLE_DRIVERS (pointed at my specific GPU's UUID) and NVIDIA_DRIVER_CAPABILITIES (set to all). Let me know if I need to make this clearer

Here's the log when playing:
API client response error: path: /Users/Me, status 400, params: {"userId":null}, data: {"type":"https://tools.ietf.org/html/rfc9110#section-15.5.1","title":"Bad Request","status":400,"traceId":"00-71c24d69ea7479d66d0f6ed59d6995cc-309f4131169d60cc-00"}, headers: {"content-type":"application/json; charset=utf-8","date":"Thu, 02 Apr 2026 13:27:31 GMT","server":"Kestrel","transfer-encoding":"chunked","vary":"Accept-Encoding","x-response-time-ms":"0.5396"}

clever topaz
#

Can you confirm that transcoding config is associated to the channel? That looks like only one line of logs. Can you post a full set of logs at DEBUG level?

sterile warren
#

you say you passed through the GPU, but did you configure Tunarr itself to use the GPU in the transcode configs?

#

sorry, should've scrolled down, let me read

#

ok so the config looks correct, like centinul mentioned we'd need to see the generated command to understand why your CPU usage is so high.

analog pier
#

I set the log level to debug, tunarr is not creating a new log since I killed and reopened the stream.

I've only got one channel running currently with only one transcode configuration still labeled Default

clever topaz
#

Start a new stream and upload the entire log. It may not create a new log file. Also, please confirm the transcode config to channel mapping I mentioned above.

analog pier
#

Accidentally deleted the channel so I remade it real quick and restarted the stream. Here are the log files.

To confirm the config mapping here is a screenshot of the config labelled Default. It's the same configuration details shown above. And a second screenshot showing "Default" as the transcode config

sterile warren
#

it is the watermark with the intermittent fade behavior that is driving your CPU usage up

#

that functionality requires CPU filters because you cannot create that effect on hardware

#

fade and / or duration (intermittent watermark) will always cause software overlay

analog pier
#

Ah, that's unfortunate. Probably not worth using those settings then

sterile warren
#

yea, a watermark without the intermittent fading can be overlaid on hardware w/ CUDA

#

i have a TODO to revisit whether that is still necessary or not (forcing software) but i havent had a chance to try it out

analog pier
#

Yep, disabled the intermittent watermark and CPU usage has gone down to expected levels. I really liked that function too...

#

Thank you both for your help solving that mystery

sterile warren