[Cin] Test rendering whit x265

Andrew Randrianasulu randrianasulu at gmail.com
Tue Jul 27 23:24:22 CEST 2021


On Tuesday, July 27, 2021, Andrea paz via Cin <cin at lists.cinelerra-gg.org>
wrote:

> I did some tests with the new and old x265. I report some data, while
> the terminal output is attached in test-x265.txt.
>
> I used a 5 min video in h264 4.2.0 but good quality. Size= 471 MB.
>
>
> x265-10bit:  CPU 50-80% (multithread)    27.5 fps    file size 46 MB
>
> x265-12bit:  CPU 50-80% (multi)          25.5 fps    file size 45 MB
>
> x265-Hi:     CPU 80-90% (multi)          10.6 fps    file size 1.7 GB
>
>
wow, this one is big!




> x265-Lo(w):  CPU 40-70% (single/multi)   34 fps      file size 67.5 MB
>
> HEVC-vaapi   CPU 0-50% (single)          85.8 fps    file size 117 MB
> (I could not see the GPU engagement)
>
> Only in x265-lo do I perceive a small decay in quality compared to the
> original. All others are comparable.
>
> For me a great news (thanks Andrew!) is the exploitation of
> multithread; on the terminal you can read the sentence:
> "Thread pool created using 16 threads"
> I have a CPU 8c/16t.


it also seems to use assembler as planned)

but a bit of concern difference between reported number of encoded frames
by x265 itself and Cinelerra:

encoded 7470 frames in 272.39s (27.42 fps), 1167.83 kb/s, Avg QP:32.61
Render::render_single: Session finished.
** rendered 7500 frames in 272.418 secs, 27.531 fps


does this mean you lost 30 frames somewhere? was this bug/difference
present in unpatched Cin?


> The only thing I regret is that the new x265 drivers introduce a CinGG
> build delay of 15 min. Before my entire compile was 5 min, now it's 20
> min.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20210728/bb82ac90/attachment-0001.htm>


More information about the Cin mailing list