I have an Nvidia GeForce 920M graphics card for notebooks. The installed driver version is 470.256.02. My operating system is Linux Mint 22 x86_64. From what I’ve seen, the maximum supported CUDA version appears to be 11.4. The problem is that on the download page for this version, the maximum supported Ubuntu version is 20.04, and the maximum supported Debian version is 10. What should I do? Am I even using the correct driver version? Can someone who really understands this topic give me guidance on the correct versions for my case? I’ve gone through several documents and forums, but I haven’t found any clear answers.
Looking at the second table under “Supported GPUs” here, your card has Compute/Cuda Capability 3.5.
Looking at the table immediately above that, the last Cuda Version supporting CC 3.5, is Cuda 11.8.
Looking at the Installation Guide for Cuda 11.8 gives supported OS’s, compilers,kernels etc.
Looking here informs that the last driver to support CC3.5, is the 470 series.
Edit: Just a note as I’m not sure what your requirements are regarding the Cuda Toolkit.
If you don’t need to compile and run Cuda based code, then all you need is the 470 driver, without any concern for OS version etc.
If you do, then attempting to install the Cuda Toolkit on an unsupported OS, may well result in install failure, that requires extra effort to achieve and may not run satisfactorily.
That’s where the problem lies: my operating system isn’t listed. I use Linux Mint 22, which is based on Ubuntu 24.04, which in turn is based on Debian 13. CUDA version 11.8 only supports up to Ubuntu 22.04 or Debian 11. Installing through the Linux Mint package isn’t an option, as it would fetch the latest versions that aren’t compatible with my machine. I even tried installing those versions with support for older versions of my OS, but without success. I noticed that the official instructions use commands that are no longer in use, such as apt-key, or something along those lines.
Unfortunately, if you are wanting to do development work, that means either changing to/dual booting a supported OS, or attempting the install via the .run installer and battling your way through install logs to find out why things failed and remedying them. Generally the problems are caused by incorrect kernel and compiler versions.
1 Like
So, you’re saying that my only options are to downgrade my operating system version, set up a dual boot, or try to install it using the .run file and manually resolve all package dependency conflicts?
Yes. Unfortunately that’s the price for the rapid advances in both hardware versions and Linux distributions.
The situation is somewhat easier for Windows users.
That’s if packages are available for the OS you’re trying to install on. It may entail compiling and installing kernels/compilers from source.
A bit sad, but it is what it is. Having CUDA isn’t a top priority. I was going to use it for Python libraries in artificial intelligence and statistics, but they also work using just the processor. I’ll just suppress the warnings and, in the future, get a more up-to-date laptop to take advantage of these features with less ‘headache.’ Thanks for the help @rs277 .
Best of luck. With due respect to possible budget constraints, I’m not sure how much benefit this GPU would be regarding AI work. It’s quite low powered and must be getting borderline to being supported in this field.
1 Like
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.