As some may know I maintain a few docker images of some available tools, and I noticed I was suddenly getting NVML mismatch, and for the life of me I could not figure out what the issue was, tried so many things, finally noticed that the docker image had some special drive 535.86.10 where my host had 535.86.05, after figuring that out I looked into it and added this to my Dockerfile:

RUN apt-get update && apt-get remove --purge -y nvidia-* && \ apt-get install -y --allow-downgrades nvidia-driver-535/jammy-updates

And voila, problem solved! Not sure what driver the docker CUDA was using, might be some special dev driver and it was causing a mismatch between the container and the host

Only started happening as of the latest driver update released late last month