I installed the docker using Compose Manager Plus. I took the opportunity to use the tuning shell scripts as supplied instead of the versions I had been using (my own code). On a like-to-like comparison, I can see that it seems to tune faster and be more stable. Some configuration tuning was required, as I am on FireTV devices with Xfinity as the source (slow tuning is always a concern).
At this point I tried turning CC's on. I was able to generate captions using CPU-only as before. When I tried using the GPU, I got the same messages as with the original docker, "cuda backend requested but not available."
I have tried referencing the Nvidia card directly, as well as trying the "--gpus=all." and get the same results either way. The container also goes into a state where no captions are generated, not even via CPU. The CC config screen then tells me to restart the container, and that starts the cycle again. For now, CC are turned off. CPU-generated CC are too resource intensive.
So, back to my original question, What can I do to triage what is going on, if I open the console while the container is running? I am familiar with the "nvidia-smi", but that does not really tell me anything that I don't already know, i.e. the container is not using the gpu.


