Posts by csaba


    Setting force_turbo=1 helped a bit, it solved the out of sync audio problem with some hevc movies.

    The gpu_mem=512 crashed the hevc movie playback with every movie I tried, so I reverted back to 320.

    I have carried out extensive tests with different downloaded hevc movies and with the Big Buck Bunny test video converted to hevc with different quality/bitrate settings. Based on the tests I think the RPI4 can play hevc videos smoothly with video bitrates below 80 Mbps, above 80 Mbps the playback stutters. Somewhere above 100 Mbps the RPI4 gets unresponsive and freezes.

    Unfortunately lots of movies are converted/remuxed with parts or the whole movie above 80 Mbps, and the RPI4’s SoC is not capable / not powerful to play them.

    I think it's more like a freeze than a crash. The RPI4 becomes so "slow", that you are not able to interact with it. You can't use the KODI UI, and you can't use the terminal to issue a command either. And after some time the system reboots. It seems to me, that it's the problem of the underlying OS and RPI4 HW.

    I will carry out more test on the weekend, and will try to find other 4K H265 video files to reproduce the problem.

    This video freezes my KODI and RPI4, sometimes the entire system reboots:


    It is referenced here in KODI wiki.

    The last log file lines on the terminal screen during freeze:

    2019-12-11 22:48:23.901 T:3011290448 DEBUG: Previous line repeats 3 times.

    2019-12-11 22:48:23.901 T:3011290448 DEBUG: CMMALRenderer::RenderUpdate - vsync 15810 (+1)

    2019-12-11 22:48:23.928 T:2627339120 DEBUG: ffmpeg[9C9A0370]: [hevc] av_rpi_zc_ref: *** Not one of our buffers: NULL

    2019-12-11 22:48:23.968 T:3011290448 DEBUG: Previous line repeats 4 times.

    2019-12-11 22:48:23.968 T:3011290448 DEBUG: CMMALRenderer::RenderUpdate - vsync 15814 (+1)

    2019-12-11 22:48:24.003 T:2627339120 DEBUG: ffmpeg[9C9A0370]: [hevc] av_rpi_zc_ref: *** Not one of our buffers: NULL

    2019-12-11 22:48:24.233 T:3011290448 DEBUG: Previous line repeats 17 times.

    2019-12-11 22:48:24.233 T:3011290448 DEBUG: CMMALRenderer::RenderUpdate - vsync 15830 (+1)

    2019-12-11 22:48:24.292 T:2627339120 DEBUG: ffmpeg[9C9A0370]: [hevc] av_rpi_zc_ref: *** Not one of our buffers: NULL

    2019-12-11 22:48:24.423 T:3011290448 DEBUG: Previous line repeats 8 times.

    2019-12-11 22:48:24.423 T:3011290448 DEBUG: CMMALRenderer::RenderUpdate - vsync 15840 (+2)

    2019-12-11 22:48:24.438 T:2627339120 DEBUG: ffmpeg[9C9A0370]: [hevc] av_rpi_zc_ref: *** Not one of our buffers: NULL


    I would like to include the LFE channel (.1) when 5.1 audio is downmixed to 2.0 by KODI. The default behaviour is that the LFE is dropped, and because of that I miss the bass effects from the movies.

    The Raspberry PI is connected to a Sony TV by HDMI, and a stereo amplifier is connected to the optical output of the TV by an optical cable. The stereo amplifier is not capable of Dolby 5.1 decoding, so the Raspberry PI + KODI has to do the job. My speakers are able to reproduce the bass of the LFE channel, but they don't get it, because the LFE is dropped during the downmix process.



    I have tested more VC-1 movies on my PI 4, and I think that if one cpu core is powerful enough to decode the movie (decoder cpu core utilization < 95%), then the playback is smooth (no frame drop), and if one the decoder core utilization is higher than 95% VC-1 playback stutters, there are frame drops.



    I have stuttering problem with some VC-1 full HD movies on PI 4 (1080p 23.976 fps 22.2 Mbps TrueHD 5.1). Player debug info reports several dropped frames, and one cpu core has 98-100% load. It can be reproduced with the VC-1 sample file on KODI Samples page: VC-1 1080i/29.970 (MKV w/DTS-HD MA 5.1 audio). Interestingly the other sample VC-1 file plays smoothly.

    9.1.002 ALPHA is installed on PI 4, 'Adjust display refresh rate' is set to Always, there is no thermal throttling (arm_freq is 1500).

    I understand, that PI 4 has no HW decode, so I can't by a licence. It seems to me, that the software decode uses only one core, and maybe it's too much work for it.

    Does anybody has the same experience?