I am late to the party but I recently came to realize that the hardware transcoding stopped working on my server back in May when I upgraded to 1.23.0 with the unable to load libcuda.so.1 error since then.
I am running the Linuxserver Plex Docker image on a QNAP server with a NVDIA card. The NVIDIA libraries used inside that container are mounted in a non-standard directory /usr/local/nvidia/lib from the parent host. Adding that path to /usr/lib/plexmediaserver/etc/ld-musl-x86_64.path resolved my issue.
I wanted to point that, up until 1.23.0, I was adding that path to /etc/ld.so.conf.d/cuda.conf and running ldconfig after that. So it seems that after 1.23.x, the PMS no longer considers the loaded shared libraries. I assume this is due to the compiler change introduced in 1.23.x?
