[FFmpeg-user] CPU load/performance change

Rhodri James rhodri at kynesim.co.uk
Fri Jul 20 14:58:39 CEST 2012

On Thu, 19 Jul 2012 19:36:33 +0100, eric cocuzza  
<ericcocuzza12 at hotmail.com> wrote:
> From: rhodri at kynesim.co.uk
>> On Thu, 19 Jul 2012 06:55:48 +0100, Carl Eugen Hoyos <cehoyos at ag.or.at>
>> wrote:
>> > eric cocuzza <ericcocuzza12 <at> hotmail.com> writes:
>> >
>> >> It takes X cycles to do do a 480P H.264 encoding/decoding then
>> >> how much it would change if the same software we
>> >> use to do 1080P H264 Encoding/decoding.
>> >
>> > I am quite sure this question cannot be answered.
>> > That said, I would expect linear performance behaviour.
>> Shouldn't it nearer to square, given that height is changing
>> as well as width?
> Square means 2-4 times right?

No, it means square.  As Carl Eugen put it elsewhere, time should be  
(roughly) proportional to the number of pixels being processed, i.e. the  
area of the screen.

Going from 480P to 1080P you are making both the width and the height of  
your movie 2.25 times bigger, so you are making the area of the movie (the  
number of pixels involved) 2.25 * 2.25 = 5.0625 times bigger.  So your  
movie will take roughly five times longer to encode or decode.

It isn't quite that simple, of course.  Dedicated hardware might scale  
better, small memory cache sizes might make things much worse, and a host  
of other factors could affect the answer.  As a rough rule of thumb, that  
ought to do, though.

Rhodri James
Kynesim Ltd

More information about the ffmpeg-user mailing list