So how much faster has x264 become since recent developments ?

[St Devious]

x264 reached revision 700 . Changelog has Altivec implementations and speedups all over it .

How much is the total speedup ?

r the speedups noticeable on all processors ?

Has Dark Shikari's speedup patches based on the test he was doing recently and other patches been added ?

More heavy development to come within the next few days ?

[burfadel]

From what I understand:
- Altivec is Macintosh, no effect on 'normal' :D computers!
- Dark Shikari's patch is the main patch (UMH, ESA) that speeds up on all processors with multiple reference frames. Not yet implemented in the SVN but a build can be downloaded.
- Speedups for recent Intel only processors

The speedups on Intel processors is noticeable, and the speedups with Dark shikari's patch is very noticeable, especially if you have 16 refs (I usually use 5 refs - the point of diminishing return for 'life' video, 16 is for animation).

[audyovydeo]

x264 reached revision 700 . Changelog has Altivec implementations and speedups all over it .

How much is the total speedup ?

r the speedups noticeable on all processors ?

Has Dark Shikari's speedup patches based on the test he was doing recently and other patches been added ?

More heavy development to come within the next few days ?


I've been asking myself these very questions : must be the 700th anniversary ;-)

Just yesterday I downloaded svn version 600 and tested against svn 697, with these surprising results (ssim, bitrate, fps) :

:: version 600
x264_0.54.600 --crf 25 -r 3 -A all -8 --direct auto -b 3 --b-pyramid -w --me hex -m 5 --bime -t 1 --mixed-refs --no-dct-decimate --no-fast-pskip --threads auto --progress --no-psnr -o NUL x_input.avs

:: 0.9984411
:: 367.9
:: 15.39


:: version 697
x264_0.57.697 --crf 25 -r 3 -A all -8 --direct auto -b 3 --b-pyramid -w --me hex -m 5 --bime -t 1 --mixed-refs --no-dct-decimate --no-fast-pskip --threads auto --progress --no-psnr -o NUL x_input.avs

:: 0.9985173
:: 388.9
:: 15.31

Clip was virtualdub's Test RGB Cube, proc was :
Intel Pentium M 745 1.80 GHz FSB 400 MHz

In the next few days I'll test on my c2d, where I expect to see significant speed differences between versions

For simplicity's sake I stick to svn, I'm not a fan of branching off a new executable every time there's a new technology. If dark shikari's new fast-ref-search needs further testing then let's get down to it and have it hit svn asap.

cheers
audyovydeo

[Shinigami-Sama]

dark shikari's isn't quite ready for SVN quite yet, with any luck it will be in a week or two from the way its been looking

and from my general lurking X264 has been multi threading and bug fixing for the last few revisions, now that most of the obvious ones are fixed we'll see how it gains speed wise

also note that theres been speed/quality improvements all over, so one resolution, and bitrate might not be effected much, but another will be