[Cin] my rendering is really slow
Andrew Randrianasulu
randrianasulu at gmail.com
Fri Jul 16 19:36:00 CEST 2021
file a bug, then... from what I remember (from manual!) yuva-8 in Cinelerra
actually more like yuv 4:4:4 (no subsampling, 8 bit resolution for each
channel) but ffmpeg should convert it automagically... more debugging
needed (
On Friday, July 16, 2021, gorge rankin <nonoobspam at gmail.com> wrote:
> Sorry, just saw your note asking me to confirm if ffmpeg can do h264/nvenc:
>
> cmd:~/cinelerra/thirdparty/ffmpeg-4.3$ ffmpeg -encoders
> Encoders:
> V..... = Video
> A..... = Audio
> S..... = Subtitle
> .F.... = Frame-level multithreading
> ..S... = Slice-level multithreading
> ...X.. = Codec is experimental
> ....B. = Supports draw_horiz_band
> .....D = Supports direct rendering method 1
>
> V..... libx264 libx264 H.264 / AVC / MPEG-4 AVC / MPEG-4 part
> 10 (codec h264)
> V..... libx264rgb libx264 H.264 / AVC / MPEG-4 AVC / MPEG-4
> part 10 RGB (codec h264)
> V..... h264_nvenc NVIDIA NVENC H.264 encoder (codec h264)
> V..... h264_omx OpenMAX IL H.264 video encoder (codec h264)
> V..... h264_qsv H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10
> (Intel Quick Sync Video acceleration) (codec h264)
> V..... h264_v4l2m2m V4L2 mem2mem H.264 encoder wrapper (codec
> h264)
> V..... h264_vaapi H.264/AVC (VAAPI) (codec h264)
> V..... nvenc NVIDIA NVENC H.264 encoder (codec h264)
> V..... nvenc_h264 NVIDIA NVENC H.264 encoder (codec h264)
>
> So yes, it's built to handle it. Also, since I'm encoing 1440p videos in
> realtime in RGBA mode, it has to be working, myh computer is not that fast
> LOL
>
> Also in the nvidia control panel, and I can see this:
> [image: image.png]
> But the percentage will hit up to 30-50% many times.
>
> On Fri, Jul 16, 2021 at 1:19 PM gorge rankin <nonoobspam at gmail.com> wrote:
>
>> oh, and I dropped down to cuda 10, that made no difference
>>
>> On Fri, Jul 16, 2021 at 1:18 PM gorge rankin <nonoobspam at gmail.com>
>> wrote:
>>
>>> some progress, So in short, it appears changing the workspace color
>>> format to YUVA-8Bit from RGBA-8Bit is where my slow down occurs.
>>>
>>> OK, so my videos are from OBS:
>>> color format: nv12
>>> color space 709
>>> color range: partial
>>> h264/nvenc/mkv
>>>
>>> so that's their defaults. those are under advanced, and I would assume
>>> normal users dont change such things.
>>>
>>> mediainfo reports the files as having yuv420 with color range of BT.709
>>>
>>> I get very fast rendering if cin has:
>>> Settings->Format RGBA-8Bit
>>> Render to yuv420
>>> Speed is 1 to 1 for the above with nvenc.
>>>
>>> However, if I use the above, then my files have too much brightness, the
>>> contrast and colors are off.
>>>
>>> So, I need to use to have my output files with proper color/bright. This
>>> is dramatically slower.
>>> Settings->Format YUVA-8Bit
>>> Render to yuv44p
>>>
>>> The only way to get my video's to look kind of like the originals in
>>> decent time
>>> Settings->Format RBGA-8Bit
>>> Render to yuv420p
>>> Add filters to adjust bright, colors. This is way faster than using
>>> YUV, but obv slower than normal RBGA as I'm using filters now which slows
>>> things down some, but at least it's liveable.
>>>
>>>
>>>
>>>
>>> On Fri, Jul 16, 2021 at 3:24 AM gorge rankin <nonoobspam at gmail.com>
>>> wrote:
>>>
>>>> ok, built it, but same issue, very slow.
>>>>
>>>> On Fri, Jul 16, 2021 at 2:45 AM gorge rankin <nonoobspam at gmail.com>
>>>> wrote:
>>>>
>>>>> I'm building cin now as I type this.
>>>>>
>>>>> I just saw this when it was building ffmpeg,
>>>>> clang: warning: Unknown CUDA version. cuda.h: CUDA_VERSION=11030.
>>>>> Assuming the latest supported version 10.1 [-Wunknown-cuda-version]
>>>>>
>>>>> Maybe my CUDA is too new?
>>>>>
>>>>> On Thu, Jul 15, 2021 at 1:26 PM Andrew Randrianasulu <
>>>>> randrianasulu at gmail.com> wrote:
>>>>>
>>>>>>
>>>>>>
>>>>>> On Thursday, July 15, 2021, gorge rankin via Cin <
>>>>>> cin at lists.cinelerra-gg.org> wrote:
>>>>>>
>>>>>>> I've only been using cin for about 3 months. I could have sworn I
>>>>>>> tested the render speed vs others and was happy. This is just odd.
>>>>>>>
>>>>>>
>>>>>> may be some library/driver was updated system-wide...?
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> I've tried the render farm idea as well. Sometimes tho, I get an
>>>>>>> empty file. Don't know why. So I'm gun shy to try it for this project
>>>>>>> where I have 3 hours estimated, it's only 15 minute video, ugh.
>>>>>>>
>>>>>>
>>>>>> yeah, emoty files for renderfarm-locally sounds like bug. I think I
>>>>>> run into some odd behavior when I was playing locally (on arm/termux) with
>>>>>> renderfarm. Try to increase number of pieces for each job and timeout, too?
>>>>>> in preferences...
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> There are no errors at all printed to the console if I launch cin
>>>>>>> from the terminal.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> odd...
>>>>>>
>>>>>>>
>>>>>>> The nvidia settings do show that about 5-8% load. I'm used to
>>>>>>> seeing that at 50-60% in other editors and also in OBS.
>>>>>>>
>>>>>>
>>>>>> what setting you have for "Use HW device" in
>>>>>> preferencies->performance? only affect decoding.. What video outputt driver
>>>>>> setting in Cin you use?
>>>>>>
>>>>>>>
>>>>>>> I know my pc is not "new" and the "fastest" but it would be faster
>>>>>>> if I were to make the compositor borderless and record that with OBS.
>>>>>>>
>>>>>>
>>>>>>
>>>>>> due to lack if hw encoding accel in termux (on android) I record with
>>>>>> system's recorder, so this is... workaround...
>>>>>>
>>>>>>>
>>>>>>> It's really odd, the compositor isn't dropping any frames
>>>>>>> whatsoever. All my effects look *awesome * in the compositor. The app is
>>>>>>> running awesome. I've not had a crash at all.
>>>>>>>
>>>>>>
>>>>>> good, but slow encoding still a problem..
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> On Thu, Jul 15, 2021 at 8:32 AM Andrea paz via Cin <
>>>>>>> cin at lists.cinelerra-gg.org> wrote:
>>>>>>>
>>>>>>>> Lately many people are experiencing performance problems in
>>>>>>>> encoding,
>>>>>>>> I am one of them. I think the cause is the lack of multithreading in
>>>>>>>> encoding (but also decoding in timeline).
>>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>> you can try to add printf("ff_cpus: %i, \n", ff_cpus) ; to few places
>>>>>> in cinelerra/ffmpeg.C and see if it was dropping down to 1 or 0 somewhat?
>>>>>> as crude hack you can force it to your number of threads....
>>>>>>
>>>>>>
>>>>>>
>>>>>> $ cat cinelerra/ffmpeg.C | grep ff_cpu
>>>>>> avctx->thread_count =
>>>>>> ffmpeg->ff_cpus();
>>>>>> ctx->thread_count = ff_cpus();
>>>>>> int FFMPEG::ff_cpus()
>>>>>> avctx->thread_count = ff_cpus();
>>>>>> $
>>>>>>
>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>> A rendering done with
>>>>>>>> external ffmpeg from command line is much more efficient than the
>>>>>>>> same
>>>>>>>> rendering done in CinGG (always using ffmpeg).
>>>>>>>> No fix has been found, only a workaround that makes use of the
>>>>>>>> Render Farm:
>>>>>>>>
>>>>>>>> https://cinelerra-gg.org/download/CinelerraGG_Manual/
>>>>>>>> Render_Farm_Usage.html
>>>>>>>>
>>>>>>>> Lately the user fary54 has made available his script to automate the
>>>>>>>> use of the render farm and external ffmpeg. Above all it is suitable
>>>>>>>> for CPUs with many threads. You can find the script and explanations
>>>>>>>> here:
>>>>>>>>
>>>>>>>> https://www.cinelerra-gg.org/bugtracker/view.php?id=575
>>>>>>>> --
>>>>>>>> Cin mailing list
>>>>>>>> Cin at lists.cinelerra-gg.org
>>>>>>>> https://lists.cinelerra-gg.org/mailman/listinfo/cin
>>>>>>>>
>>>>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20210716/f9ac8c27/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 9735 bytes
Desc: not available
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20210716/f9ac8c27/attachment-0001.png>
More information about the Cin
mailing list