Error: "Existing package manager installation of the driver found..."

When attempting to install Cuda Toolkit (Cuda_10.1.243_418.87.00) under Linux Mint Cinnamon, I get the following error message:

" Existing package manager installation of the driver found. It is strongly
recommended that you remove this before continuing.
Abort
Continue
"

I’ve read of some serious problems that have occurred when uninstalling Nvidia graphics drivers (like complete system crashes). So of course I chickened out. :-)

What is the recommended procedure here? I’m installing the Cuda Toolkit for neural net programming, if that matters.

Linux Mint isn’t a supported distro for CUDA.

The recommended procedures in general are covered in the linux install guide, particular the section on “handle conflicting installs”

Hi Robert,

Mint is based on Ubuntu, so I wouldn’t expect any problems with compatibility. But I believe the driver-removal disasters that I’ve read about were primarily Ubuntu systems.

In general, are there likely to be problems restarting the system if the NVidia driver is uninstalled (even if the intent is to replace it immediately)? The tales that I’ve heard involve complete loss of video and UI. It would be great to know that there is a procedure for avoiding that.

Mark

If you switch to runlevel 3, you will certainly not lose video or UI. The UI in this case is console/text. On ubuntu it would be roughly equivalent to disable the gdm/lightdm. In the case of a non-functional gdm, on ubuntu you can I believe insert commands on startup to disable loading of the gdm, and/or you can login remotely over the network, and generally fix things from there. I personally would not work on an ubuntu machine until I had enabled SSH for network logins.

I haven’t heard those tales. I’m unfamiliar with that case. The primary problem I’ve heard with ubuntu is the login loop. It occurs on driver install, and uninstall generally fixes it according to what I have read. If you want to find out more about that, a google search on “ubuntu login loop” will turn up examples.

Thanks for your reply, Robert. I’ve been traveling but I’m back to the grind. :)

I have read through the “NVidia Cuda Installation Guide for Linux.” I looked for
the ‘conflicting installs’ section that you mentioned but did not get fine nuance
for all of it.

After some searching, I had initially installed the 1080Ti GPU, then blacklisted
Nouveau, then installed the Nvidia driver first (without the complete toolkit).
In retrospect, maybe I should have just tried to install the complete toolkit
via the .run file to begin with instead of just the driver. Perhaps that is the
current hitch.

I suppose I could try to back out of that installation process by un-blacklisting
Nouveau and attempting to uninstall the driver, etc. but that seems tedious and
error-prone.

Given that the NVidia driver seems to be working, can you think of any reliable way
to install the remainder of the Cuda Toolkit while keeping the driver intact?

I’ve seen recommendations for installing via “sudo apt install nvidia-cuda-toolkit”
but I thought I should check back with you before attempting that. I don’t want
to induce more problems.

Hi, i got the same cas : i’m with Ubuntu 18 ; i switched to ‘init 3’ level, i did remove nvidia drivers from the kernel and i continue having that stupid message, is that possible i got this because i connect to the system remotely with x2go ?