GPU usage is too high during hardware transcoding

[Peaynan]

When I use Jellyfin for hardware transcoding, the GPU usage is very high, reaching up to 95% or more.

However, in a previous version (I can't recall which one), the GPU usage remained stable at around 30%.

I've noticed that the transcoding logic seems to quickly transcode the set portion (I've set it to 3 minutes) by utilizing extremely high GPU usage, and after the task is complete, the GPU usage drops back to 0. This causes a problem: when someone is streaming with transcoding, my machine generates a lot of noise. I would prefer my GPU to continuously transcode rather than finishing the task quickly and then stopping.
（This is unrelated to the video bitrate, as this situation can occur even with a very low bitrate.）

My CPU is AMD Ryzen 7 8845H, and my GPU is RTX 4060 Laptop.

I am not sure if this is a bug or a change in the new version. I hope someone can clarify my doubts, and it would be great if a solution could be proposed. Thank you.

[Fate]

I'm confused....
Is the server running on the Laptop? If yes, with what device are you watching?

You could try to enable the "throttle transcoding" option in jellyfin, then jellyfin will only transcode 3mins of video and then wait for the viewer to catch up.

Also transcoding should not let your GPU-fan to be much louder since encoding takes place on a special area of the chip that does not generate as much heat as a game for example.

Can you post a transcode log file?

[TheDreadPirate]

Other than what Fate mentioned, "throttle transcoding", there is no mechanism available in ffmpeg to only go "50%". ffmpeg will always go as fast as possible. Throttle transcoding will pause ffmpeg, but will still run as fast as possible in between pauses.

Since the compute usage is also high, I am assuming this was an HDR video that was tone mapped. Which would use more power than an SDR transcode.

[nyanmisaka]

For NVIDIA, all video processing filters are executed on GPU/CUDA, not in fixed-function hardware like Intel's VPP. This means more power consumption and more heat generation compared to Intel iGPU. And higher GPU utilization means there are almost no bottlenecks in the transcoding pipeline, which is good for video transcoding performance.

If you don't want fan speed fluctuations due to transcoding pauses and continues, you can disable the "throttle transcoding" option in "Dashboard->Playback".