jetson-ffmpeg VS voukoder

I made my own build of frigate using the community nvmpi ffmpeg decoder. What I didn't realise when I bought Jetson is normal Nvidia nvdec ffmpeg is not supported on Jetsons, and nvidia did little to enable their different hardware decoder for Jetsons, and concentrated their efforts on gstreamer instead of ffmpeg. So you need to use 3rd party nvmpi extension for ffmpeg https://github.com/jocover/jetson-ffmpeg . Also I debloat Frigate ffmpeg build. I spent a lot of time trying to get the hardware decoder working on rk3399 before giving up and getting a jetson nano. Reducing build time by removing all the unnecessary codecs is helpful for testing. https://pastebin.com/bxWwDz0K is my ffmpeg config for Frigate. Copy and create a new build in Frigate Makefile for aarch64nvmpi based off the aarch64 config. Make that config use the ffmpeg config specific to nvmpi.

I purchased the jetson nano because the initial specs indicated that real time encoding for 1080p 60fps was possible. Later did I figure out that the jetson nano has a different driver for their onboard gpu different than the desktop gpus. The desktop gpu use nvenc dedicated hardware for encoding that works with ffmpeg, alas for the jetson nano, nvidia does not support ffmpeg hardware accelerated encoding out of the box, only decoding. But, someone actually came up with a solution to include the nvmpi lib that utilizes the nvenc hardware acceleration for encoding. https://github.com/jocover/jetson-ffmpeg

Also voukoder is really recommended as it uses ffmpeg to render and it's far superior to adobe's integrated rendering stuff. This will use your gpu completely without limits if you choose render with it. https://github.com/Vouk/voukoder

Default x264 from Adobe does this, I recommend using Vuokoder (which it usees FFMPEG) as a ground for better quality. https://github.com/Vouk/voukoder

You could try Voukoder