On Tue, Jul 02, 2019 at 02:47:35PM +0200, Peter Corlett wrote: > It's debatable whether a single ffmpeg instance could take advantage of that > many cores since Amdahl's Law will kick in as it tries to co-ordinate > everything. Split it into multiple four- or eight-thread encodes and run them > in parallel on that monster server, or even better, run them in parallel on a > fleet of much-cheaper desktop machines.
>From my understanding of how yer typical video codec works (which could of course be wrong!) I would think it's one of the few common tasks that can take advantage of that many CPUs, as the parallelisable proportion is very large. A video file consists of a list of chunks, each of which consists of one complete frame followed by a bunch of diffs from one frame to the next. If you're decoding, hand one chunk to each CPU and process them in parallel, and hand each CPU a new task when it finishes. There's some small overhead in figuring out where each chunk begins, and in wrangling pointers so that you end up with the results in a sequence of decoded frames. Encoding is of course similar. -- David Cantrell | top google result for "internet beard fetish club" I'm in retox _______________________________________________ get_iplayer mailing list get_iplayer@lists.infradead.org http://lists.infradead.org/mailman/listinfo/get_iplayer