[General] Roku and or Chapter extraction FFMpeg

[ThermoDust 12]

I do not know which process this is coming form either the Roku plugin or the chapter extraction... but I was wonder if it is possible or if there already is a flag to make it so this process could use Intel Quicksync / Hardware Acceleration?

 

20952 emby       20   0  717M 67432 19744 S 357.  1.7 11:42.61 │     ├─ /usr/bin/ffmpeg -i file:/media/4207a26e-560e-4e16-9b66-094e59fe6960/TV Shows/Continuum/Season 03/Continuum - S03E09.mp4 -thre

20962 emby       20   0  717M 67432 19744 R 70.8  1.7  2:18.62 │     │  ├─ /usr/bin/ffmpeg -i file:/media/4207a26e-560e-4e16-9b66-094e59fe6960/TV Shows/Continuum/Season 03/Continuum - S03E09.mp4 -t

20961 emby       20   0  717M 67432 19744 R 69.7  1.7  2:18.28 │     │  ├─ /usr/bin/ffmpeg -i file:/media/4207a26e-560e-4e16-9b66-094e59fe6960/TV Shows/Continuum/Season 03/Continuum - S03E09.mp4 -t

20960 emby       20   0  717M 67432 19744 S 71.0  1.7  2:18.46 │     │  ├─ /usr/bin/ffmpeg -i file:/media/4207a26e-560e-4e16-9b66-094e59fe6960/TV Shows/Continuum/Season 03/Continuum - S03E09.mp4 -t

20959 emby       20   0  717M 67432 19744 R 69.9  1.7  2:18.22 │     │  ├─ /usr/bin/ffmpeg -i file:/media/4207a26e-560e-4e16-9b66-094e59fe6960/TV Shows/Continuum/Season 03/Continuum - S03E09.mp4 -t

[Luke 37248]

It's a possibility for the future, yes.

[Waldonnis 148]

 

I do not know which process this is coming form either the Roku plugin or the chapter extraction... but I was wonder if it is possible or if there already is a flag to make it so this process could use Intel Quicksync / Hardware Acceleration?

 

I'm not sure using hardware decoding is going to help in the way you might expect for Roku thumbnail generation unless you're fine with the ffmpeg image extraction being single-threaded (hardware decoding and multi-threading don't mix well).  I ran some tests after seeing this message and found that while my CPU usage was significantly less, it also took a bit longer due to that limitation.  For background bif generation, this may not be a bad thing, though.  I've already altered my personal script to use dxva2 decoding, actually, since my computer is usually busy and I don't care if it takes longer.  To give you an idea, normal software decoding was saturating the CPU, while swapping to single-threaded hardware decoding dropped CPU use to ~11-15%.

 

For the life of me I couldn't get cuvid to work, and I've had problems with QuickSync decoding recently (driver related, but I haven't tracked down which driver version introduced the artifacting that I'm seeing), so I just used dxva2 (Windows-only;I would assume VA-API could be used on the Linux side, but haven't tried it).

 

I also tried a short transcode using the h264_nvenc encoder and dxva2 for the decoding and while CPU use was lower (83% vs 20%, it didn't show any appreciable speed benefit.  Admittedly it was a short clip, so it may show some more notable speed gains/losses if I let it run against a longer clip.

Edited January 26, 2017 by Waldonnis