What version of Pytorch is compatible with NVIDIA GeForce GTX 1080

1

I updated my Pytorch installation last night, but now I am getting a message:

UserWarning:
Found GPU1 NVIDIA GeForce GTX 1080 which is of cuda capability 6.1.
PyTorch no longer supports this GPU because it is too old.
The minimum cuda capability supported by this library is 7.5.

My current Cuda version is 12.9. I am running Ubuntu 24.04.

I checked Chatgpt and it was telling me all sorts of things about using using CUDA 11.8 and reducing the versions for a bunch of different packages. That sounded a bit suspicious to me, so I figured I would check.

I consulted the documentation on the pytorch website, but all of the different versions of cuda, and pytorch, and cudatoolkit, etc., are confusing me. And I am worried about doing major surgery on my cuda installations and then putting that whole ecosystem into an unrecoverable state.

Can someone indicate what the correct pytorch version is for a GTX 1080 now? I imagine pytorch 2.5 works with cuda 12.9, as that is what I had before. Sorry if this question is too direct, however I did consult multiple sources before finally posting here. Any suggestions are appreciated.

2

Install either any older binaries (<=2.7.1) or our current binaries built with CUDA 12.6.

1 Like
3

Oh yes, this is exactly what I was looking for. Thanks so much @ptrblck . I get it.

So I just have to install the nvidia-cuda-toolkit that corresponds to CUDA 12.6. Meaning I unintall 12.9 and then install 12.6. Then install pytorch 2.7, and then things should work.

Let me know if I captured anything incorrectly. Otherwise I will proceed down that path. Thanks again for setting things straight.

4

You don’t need to install a CUDA toolkit yourself unless you want to build PyTorch from source or a custom CUDA extension (or any other CUDA application of course). The PyTorch binaries ship with their own CUDA runtime dependencies and you would only need to properly install an NVIDIA driver from the same major release.

2 Likes