Pi4, hardware acceleration, fast forward freezes

mooncaptain · April 30, 2021 at 5:44 PM

I am running a Pi4 with 2GB of ram. SD card for OS. Emby server + nas for data storage, Pi4 is mounted in a Flirc all aluminum case. No fan.

I was having a display freeze problem when I used fast forward during video playback. The whole display would lock up but I could still access the system with SSH. Still required a reboot to get control through the user interface.

I turned off Allow Hardware acceleration MMAL and then fast forwarding worked just fine.

While using fast forward debug is on. I see CPU usage on one of the CPU's runs at 100% while the other 3 are in the teens or lower. After some seconds the high activity moves to a different CPU. I suppose this is happening to manage heat dissipation.

stepping through replay speeds (1x,2x...32x) on the 1080p video I see a range of cputemp readings from 40c to 46c.

All this is really great. I can't imagine that I would see a better heat profile even if MMAL worked so I will leave hardware acceleration off.

**Da Flex** · April 30, 2021 at 5:59 PM Official Post

RPi4 can hardware-decode some codecs. If you re-encode your videos to one of those, you might have a benefit in spooling.

**Da Flex** · May 1, 2021 at 7:51 PM Official Post

To quote RPi4's tech specs:

Quote

H.265 (4kp60 decode), H264 (1080p60 decode, 1080p30 encode)

So re-encode to H264 or H265, and you can expect better spooling behavior.

mooncaptain · May 2, 2021 at 4:16 AM

Thank you for the information. I do have x264 and h264 encoded files that MMAL does support but these files have the problem with fast forwarding. Since I have found a solution i'll stick with that.

Thanks again

**Da Flex** · May 2, 2021 at 4:47 AM Official Post

Because H264 / H265 decoding is already hardware accelerated, activating MMAL makes no sense to me.

I think your RPi4 works as it should, if that's what you're asking for.

**chewitt** · May 2, 2021 at 5:54 AM Official Post

LE 9.2 reached a point where it worked for the majority of use-cases, and then the focus moved away from legacy MMAL/OMX decoding onto the all-new GBM/V4L2 codebase. The older methods still have rough edges that will never be fixed, so if you have problem media I'd suggest you retest under LE10, because bugs/issues found there are of interest to the developers.