So Ubuntu 26.04 is out and it promises to come with all the Nvidia drivers, cuda and pytorch all out of one distribution. Great! For the last 7 or more years I’ve used lambda labs, I’d like to remove that dependency.
So I install ubuntu-26.04-live-server-amd64.iso and select non-free drivers. Once that is up and running Iapt install cuda-toolkit nvidia-cuda-toolkit python3-torch nvidia-utils-595-serverand great, nvidia-smi can see my GPUs. But pytorch can’t see my GPUs:
tonyr@ml0:~$ python3
Python 3.14.4 (main, Apr 8 2026, 04:02:31) [GCC 15.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
False
I don’t know who is building python3-torch but it’s nobody from the core or module maintainers.
Thanks. I wasn’t trying to file a bug report, I was trying to reach out to the pytorch community. I can believe that I’m missing one step between having pytorch running and it seeing my GPUs. It’s probably not a problem with whoever built python3-torch for Ubuntu 26.04, but everyone here is running pytorch and many will be using Ubuntu, so this seemed like a good place to find someone who had got this working. Also now posted at https://askubuntu.com/questions/1566422/
I believe it is. A quick search for this package shows it’s an old PyTorch 2.9.1 release without any indication of CUDA support: link which is why I asked where the CUDA support was claimed.
I didn’t see it as a bug report and yes, thanks for reaching out here.
Our general recommendation is to use our wheels from the install matrix and to report any issues you might be hitting.
And you were right - yet again - thanks a lot. I’m not the only person who is going to have this problem, so let me write it up and hopefully save others a lot of effort.