[theora] codec efficiency? was: is this FUD or not?

[xiphmont at xiph.org]

On Wed, Mar 24, 2010 at 3:03 AM, startx <startx at plentyfact.org> wrote:
>
>> On Wed, Mar 24, 2010 at 03:37, Tom Sparks <tom_a_sparks at yahoo.com.au>
>> wrote:
>
>> > here is the message:
>> > "Intrinsically H.264 properly encoded should be about 30% more
>> > efficient than Theora."

The exact number is almost certainly wrong, for one thing it's too
round :-)  We can say with certainty that the best h264 encoders (eg,
x264), when allowed to use High profile, are currently far more
efficient than Theora and that includes Ptalarbvorm.  The price of
being 'better' is being slower and requiring more memory to both
encode and decode (roughly factor of 5-10 I believe).  The High
profile encodes are usable on full-blown computers, but generally
don't play on mobiles or less powerful devices, even with hardware
assist.

Using h264 Baseline (what is actually universally implemented) the
bitrate efficiency difference mostly disappears.

Also, we're not currently interested in extending the bitstream to
'beef it up' as tempting as that may be.  It actually has a reasonably
nice spot on the performance/resources curve and stability (that is,
continued compatability) is far far more important than chucking
compatability to close a 30% gap.  Theora may be an out of date format
(and in some ways it is--- though not *nearly* as out of date as, say,
mp3) but we still have room to grow in it.  As old as the format is,
it's still not being fully exploited, though we're getting closer to
that point.

> i have a rather stupid question: is there an agreed interpretation of
> what people generally mean when they say a codec is more "efficient"
> then others?

Oh sure.  It's virtually always taken to mean 'better perceptual
quality at a given equal bitrate'.  It's valid to take resources
required into account, but I wouldn't assume anyone did unless they
explicitly said so.

>
> does this mean:
>
> a) the resulting file (in same quality) of one encoder is smaller then
> the others?
>
> b) the encoding process is quicker?

People nearly always mean a).  If it was b), there'd be no question
Theora was very efficient.

Monty