[MEncoder-users] Performance on multi core machines - a specific enquiry re avisynth mt input

Michael Rozdoba mroz at ukgateway.net
Sat Sep 29 23:48:51 CEST 2007


RC wrote:
> On Sat, 29 Sep 2007 21:18:24 +0100
> Michael Rozdoba <mroz at ukgateway.net> wrote:
> 
>> perhaps MEncoder is still only allowing the decoding of its input to
>> run in a single thread.  Is this possible?
> 
> Mencoder only uses a single thread, unless you tell it otherwise (and
> even then it's highly dependant on the codec and input).

I do tell it to use multiple threads, but suspect this only affects the 
encoding process, not also decoding of the input.

> Decoding should never be a bottleneck, unless you've got 16+ cores, or
> some other very unusual scenario.

Four cores, but the reason decoding /is/ a bottleneck in this case is 
that it involves frameserving via a complex script into MEncoder.

Can you confirm that decoding of the input will only use a single thread?

As I said in another post, I don't understand Windows internals, but it 
does seem quite clear that while decoding of this script can & does use 
multiple threads in most cases, when used as input to MEncoder, it is 
confined to a single thread.

FWIW this is an important issue in Megui afaic since the raison d'etre 
for including MEncoder, aiui, is purely to perform the huffy encode for 
a pre-rendering step which is only needed when the input avs is much 
slower to decode than the encoding passes are to process (in a multi 
pass encode).

If the prerendering step runs at much below 50% of what it would if done 
concurrently with the encode, it's obviously a waste of time performing 
it as a distinct stage.

-- 
Michael Rozdoba



More information about the MEncoder-users mailing list