this post was submitted on 13 Dec 2023
125 points (99.2% liked)

Jellyfin: The Free Software Media System

5741 readers
68 users here now

Current stable release: 10.10.2

Community Standards

Website

Forum

GitHub

Documentation

Feature Requests

Matrix (General Information & Help)

Matrix (Announcements)

Matrix (General Development)

Matrix (Off-Topic) - Come get to know the team and blow off steam!

Matrix Space - List of all the available rooms on Matrix.

Discord - Bridged to our Matrix rooms

founded 4 years ago
MODERATORS
 

This should eventually make it's way into jellyfin. Eager to see the performance improvements.

top 6 comments
sorted by: hot top controversial new old
[–] [email protected] 37 points 11 months ago (2 children)

As far as I'm aware this is only for the cli version of ffmpeg and won't affect the threading of codecs many of which were already multithreaded.

[–] Lmaydev 7 points 11 months ago

It's literally for the CLI yeah.

[–] [email protected] 2 points 11 months ago (2 children)

I recently tried figuring out how to build ffmpeg with Nvidia codecs. I'm very new to ffmpeg and codec terminology. How is multithreading for the CLI different than the codecs?

[–] [email protected] 7 points 11 months ago

My understanding is some parts have to be done sequentially even though the parts themselves are multithreaded, now the different parts can all be done in parallel.

[–] Lmaydev 1 points 11 months ago* (last edited 11 months ago)

Change the main loop and every component (demuxers, decoders, filters, encoders, muxers) to use the previously added transcode scheduler.

The components are what they have parallelised.

[–] [email protected] 4 points 11 months ago

This is the best summary I could come up with:


The long-in-development work for a fully-functional multi-threaded FFmpeg command line has been merged!

FFmpeg is widely-used throughout many industries for video transcoding and in today's many-core world this is a terrific improvement for this key open-source project.

The patches include adding the thread-aware transcode scheduling infrastructure, moving encoding to a separate thread, and various other low-level changes.

Change the main loop and every component (demuxers, decoders, filters, encoders, muxers) to use the previously added transcode scheduler.

There's a recent presentation on this work by developer Anton Khirnov.

It's terrific seeing this merged and will be interesting to see the performance impact in practice.


The original article contains 226 words, the summary contains 103 words. Saved 54%. I'm a bot and I'm open source!