x264 build 596 speedup

[pyrates]

anyone notice running build 596 during the second pass that the time is the same as the first pass?

Here is the options I use when doing the first pass to generate the stats file:

"D:\Program Files (x86)\x264\x264.exe" --pass 1 --bitrate 3958 --stats "N:\movie\1080i.stats" --bframes 3 --b-pyramid --direct auto --filter -2,-1 --subme 1 --analyse none --vbv-maxrate 25000 --me dia --threads 2 --cqmfile "N:\eqm_avc_hr.cfg" --progress --no-psnr --output NUL "N:\movie\1080i.avs"

And here is the options I use for the 2nd pass:

"D:\Program Files (x86)\x264\x264.exe" --pass 2 --bitrate 3958 --stats "N:\movie\1080i.stats" --ref 16 --mixed-refs --no-fast-pskip --bframes 3 --b-pyramid --b-rdo --bime --weightb --direct auto --filter -2,-1 --subme 7 --trellis 2 --analyse all --8x8dct --vbv-maxrate 25000 --me umh --threads 2 --cqmfile "N:\eqm_avc_hr.cfg" --progress --no-psnr --output "N:\movie\1080i.mp4" "N:\movie\1080i.avs"

Or is this a bug I just found?

[DarkZell666]

who's build was it ? b0bor's from x264.nl ? Sharktooth's ? an other one ?

Edit: obviously not Sharktooth's :)

rev 596 says: compute chroma_qp only once per mb
Will try right now ...

[DarkZell666]

negative, using the same options (rev 596 from x264.nl) I get 26fps on first pass, and less than 3fps on second pass.

note: setting bitrate to 500 and using trellis 1 I get 40fps on 1st pass (waiting for 2nd ... *edit: less than 5fps*)

What's your system's config ? Can you reproduce this "symptom" more than once, using different sources and settings ?

[pyrates]

ah my mistake. Only showed up for like half an hour then went back down to the same speed as the previous build.

[Adub]

Damn!
And you went and got my hopes up too.

[burfadel]

Well adding the command for 16 reference frames, RDO on B frame, a subme of 7, a trellis set to always etc is the reason! The massive slowdown is expected. Try using:
5 reference frames
subme of 6
trellis of 1

leave the RDO on b-frames, but it does slow it from your first pass!

And the speed will be much more respectable! You wouldn't even notice a quality drop, if you do then you're too fussy! :D hehe, nah seriously, those settings you chose are in the ultra insane category, even the ones I suggested is way above the quality that is standard.

Ideally for the first pass a different analysis engine would be greatly beneficial as it could be undertaken at much greater speeds, but this would require lots of extra code and programming knowledge which isn't practical.

[DarkZell666]

Sorry mate, you got it the wrong way round :)

pyrates thought that the 2nd pass was as FAST as the first pass (hence the topic title). He wasn't complaining about the slowness of x264 but it's apparent ultra-fastness (which, in the end, was too quick of a conclusion, as he stated afterwards ;))