I am finally making progress with setting up an PeerTube instance but it turns out to be quite complex with object storage and all that.

The latest issue I am facing is that the official Peertube containers do not have any GPU hardware acceleration support for transcoding or Whisper subtitle generation.

But instead of trying to monkey-patch those official containers, I assume it would be smarter to run a separate container with the peertube-runner process and the necessary NVENC / VAAPI and ctranslate2 libraries installed?

But to my surprise there seems to be no suitable container to be easily found that covers all that, in fact people even deprechiated NVENC plugins because of bad quality of the transcoding?

I find this a bit confusing, because especially for livestreaming HW accellerated transcoding really seems much superior.

Anyone has a good suggestion where I could maybe find such a container or a guide how to set this up? Thx!

  • poVoq@slrpnk.netOP
    3·
    3 days ago

    OK, found one that does at least the Whisper transcription, and might be a good basis for NVENC as well with some tweaks.

    • poVoq@slrpnk.netOP
      2·
      3 days ago

      Thanks. That links does have some good workarounds/hints on how to maybe make a suitable container.

  • Meldrik@lemmy.wtfM
    1·
    3 days ago

    You can still try to install the NVENC plugin and see if works. I’ve used it before.

    • poVoq@slrpnk.netOP
      3·
      3 days ago

      It would probably work if I wasn’t running it in the official docker containers. I tried it with the Nvidia GPU passthrough to container stuff (which works when containers are meant to be used with CUDA etc), but the official Peertube container itself just lacks the necessary software inside to make use of the GPU I think.