[FFmpeg-devel] Looking for a consultant

[Carl Eugen Hoyos]

2018-06-20 16:50 GMT+02:00, Ariel Frailich <ariel at websiteatelier.com>:

> Input: mostly ProRes (10-20Gb), HD to 4K res, 90 mins average length.
>
> Output: a set of .mp4 files at 5 different resolutions, up to 1080.
> These are further processed (primarily bento4) to create multiple/
> variable bitrate HLS and DASH streams, with encryption and DRM.

(The most important information would be the target codec.)

> On our office machine (iMac, 3.2 GHz Intel Core i5, 8Gb, macOS
> Sierra), encoding typically takes 6-7 hours per film. Ideally, we
> would like to cut this down to 1-2 hours.
>
> WHAT WE NEED TO KNOW
>
> 1. Hardware: what's a good hardware configuration for our needs?

That depends on the question if you want hardware or software
encoding (that nobody here can answer, only you).

> CPUs, cores, RAM, RAM disk, graphics cards, etc.

The more cpu cores the faster software encoding is, you
will find information online about the performance of
graphic cards for hardware encoding.

> Also, which version of Linux is recommended (or which to avoid).

No limitations known.
(musl 32bit is unsupported which we are not allowed to
document but this will not hit you.)

> 2. ffmpeg setup: how should ffmpeg be compiled to best take
> advantage of the particular hardware configuration?

$ ./configure --enable-gpl && make for hardware encoding,
$ ./configure --enable-gpl --enable-libx264 && make for
software encoding.

> 3. Video processing: how should ffmpeg be run to maximize
> speed and efficiency, and to produce the most suitable files
> for our needs?

I may misunderstand (and this is possibly not an FFmpeg-
related question) but for software encoding, remember that
threads do not scale well, single-threaded encoding (of
multiple files at the same time) is always most efficient.

Carl Eugen

PS: In case you don't know: Hardware encoding is potentially
faster than software encoding but provides worse quality for
a given bitrate or larger files for a given quality.