vp9 --cpu-used options

[ppp....@gmail.com]

Hi, VP9 community,

Recently, I'm testing the VP9 encoding performance. Specifically I tried to adjust the quality vs. the speed tradeoff, i.e. setting --best, --good, or --rt combined with --cpu-used option. I would like to know whether the command line documentation for VP8 at http://www.webmproject.org/docs/encoder-parameters/  is still valid for VP9. Sorry I cannot find the office documentation for VP9 command line options. It seems that for VP9, --cpu-used can range from -8 to +8,  are all those steps available for both --good and --rt options, or -ggod and --rt use different step values and their meaning is different? Any explanation is greatly appreciated.

Thanks,

Peng.

[James Zern]

Hi,

On Sat, Aug 27, 2016 at 10:27 AM, <ppp....@gmail.com> wrote:
> Hi, VP9 community,
>
> Recently, I'm testing the VP9 encoding performance. Specifically I tried to
> adjust the quality vs. the speed tradeoff, i.e. setting --best, --good, or
> --rt combined with --cpu-used option. I would like to know whether the
> command line documentation for VP8 at
> http://www.webmproject.org/docs/encoder-parameters/ is still valid for VP9.
>

The descriptions in general are applicable; you can substitute
--tile-columns for --token-parts for VP9. It would also be best to
stick to --good and --rt, --best doesn't get much attention and will
likely be too slow to be usable.

> Sorry I cannot find the office documentation for VP9 command line options.
> It seems that for VP9, --cpu-used can range from -8 to +8, are all those
> steps available for both --good and --rt options, or -ggod and --rt use
> different step values and their meaning is different? Any explanation is
> greatly appreciated.
>

For vbr/cbr/cq 2-pass you can use cpu-used=4 for the first pass then
try 0/1/2 for the second pass (for a 1-pass encode use 0/1/2). With
--rt try 5-8, 5 will be best on a desktop, while on a mobile platform
you may want to start at 8.
To start with you may want to work with vbr/cbr and --target-bitrate,
--min-q, --max-q and --kf-max-dist to get reasonable output.