I have a few servers with Tesla T4 GPUs in them. The T4s are older GPUs, roughly 75% as powerful as the venerable GTX 1080. I am thinking that they would be fairly suitable to put to work doing hardware transcoding as Remote Runner servers.
However, I am not sure if the Tesla T4 is compatible with Peertube in this capacity (hardware transcoding in remote runners).
Is there a list of compatible GPUs for hardware transcoding that I could look up, and see if the Tesla T4 is on it? Or, to handle hardware encoding, do I need some special version of ffmpeg that was compiled with CUDA, NVENC, NVDEC, etc., as per Nvidia’s page on the subject?
And… if I actually DO need a special build of ffmpeg, is there some way to have peertube insert whatever special flags might be needed to have ffmpeg utilize the hardware? Or is there a way to compile it so it does that automatically, and Peertube wouldn’t need to worry about it?
Thanks, I reviewed the linked topic. To judge from the responses there, it sounds like you will eventually be incorporating hardware acceleration for transcoding (both in the local Peertube Server as well as Remote Runner Servers) at some future date, but for the time being, your development schedule is already planned out to the end of the year.
Hope all is well on your end. I am circling back now as we approach the middle of 2026, as I have heard that Remote Runners now do support CUDA-based Transcoding.
I currently have a Dual-CPU 88-Core Server running Ubuntu Linux with four 16 GB Tesla P100 GPUs and 200 GB of RAM in it that I want to use for CUDA-accelerated transcoding of multiple live video streams.
How do I set up four Remote Runner instances for this purpose on the Server, each pointed at a different GPU (0-3)?
Is it a modification of the config.toml file so that the remote runner(s) would automatically distribute jobs across the available gpus in the system, or would it be a flag on the command that runs each remote runner, like “–gpu 0” or something similar? I haven’t found any reference to it.
