x264 option analysis

[ronny85]

given this partial set of options for x264 cli


--ref 9 --deblock -3:-3 --8x8dct --partitions all --me umh --subme 10 --psy-rd 1.00:0.15 --mixed-refs --merange 32 --trellis 2 --qpstep 4 --deadzone-inter 21 --deadzone-intra 11 --rc-lookahead 60 --cqm flat --scenecut 40 --qblur 0.5 --keyint 240 --min-keyint 23 --qcomp 0.60 --ratetol 1.0 --aq-mode 1 --aq-strength 1.00 --ipratio 1.40 --nr 0 --b-adapt 2 --weightp 2 --chroma-qp-offset -3 --cplxblur 20.0 --b-bias 0 --nal-hrd none --bframes 12 --b-pyramid normal --no-fast-pskip --open-gop


which one is the most computing intensive (slows down the most)?

[microchip8]

9 reference frames will considerably slow down things, and is most of the time very little beneficial for live action content. Good value is between 4 and 6. In terms of most CPU intensive, an encoder spends most of its time in motion estimation which is controlled by more than one option like --me and --subme and --merange, etc

[ajp_anton]

Also IIRC, --b-adapt 2 together with a high number of --bframes also slows things down (mainly for being single-threaded?).

[benwaggoner]

Also IIRC, --b-adapt 2 together with a high number of --bframes also slows things down (mainly for being single-threaded?).
Also, there was a patch some months ago that was supposed to make --b-adapt 1 about as good as --b-adapt 2. If it worked reasonably well, the default and presets <placebo should probably switch to using --b-adapt 1.

I haven't exhaustively tested it myself, though.

[sneaker_ger]

Preset slow was subsequently changed from --b-adapt 2 to 1. Only slower to placebo use 2.