I have Ubuntu 22.04 with Nvidia 515.86.01 (proprietary), along with the CUDA toolkit and cuDNN.
nvidia/515.86.01, 5.15.0-53-generic, x86_64
After 3 weeks of holiday I am back and wanted to install a few tools (Evince for example). I was surprised to see that nothing regarding apt-get related to package installation or upgrade works due to a mysterious libnvidia-nscq:
Reading package lists... Done Building dependency tree... Done Reading
state information... Done You might want to run 'apt --fix-broken
install' to correct these. The following packages have unmet
dependencies: libnvidia-nscq-510 : Depends: libnvidia-nscq-515 but it
is not installed E: Unmet dependencies. Try 'apt --fix-broken install'
with no packages (or specify a solution).
From what I can tell
- NSCQ is a NVswitch, which is something from Nvidia regarding servers and GPUs
- Version 510 is a transitional package
Running dpkg -l *nvidia* returns the following (among others)
iU libnvidia-nscq-510 515.86.01-0ubuntu0.22.04.1 amd64 Transitional package for libnvidia-nscq-515
in libnvidia-nscq-515 amd64 (no description available)
I cannot figure out where this NSCQ dependency is coming from. In addition it is rather strange that 510 depends on 515 but perhaps I've misunderstood this line.
I can neither run apt-get upgrade, nor apt-get autoremove, nor apt-get install <package>. I did apt-get clean to remove the cached packages and then pulled fresh ones using apt-get update.
If I am to remove the Nvidia drivers and CUDA toolkit, I am pretty sure it will break my machine learning setup (PyTorch and TensorFlow). These tools are very fiddly when it comes to which version of driver/CUDA/cuDNN is to be used.