Package Details: cuda-12.9 12.9.1-1

Git Clone URL: https://aur.archlinux.org/cuda-12.9.git (read-only, click to copy)
Package Base: cuda-12.9
Description: NVIDIA's GPU programming toolkit version 12.9 (Maxwell/Pascal/Volta support)
Upstream URL: https://developer.nvidia.com/cuda-zone
Licenses: custom:NVIDIA
Conflicts: cuda
Provides: cuda, cuda-sdk, cuda-toolkit
Replaces: cuda-sdk, cuda-toolkit
Submitter: piernov
Maintainer: piernov
Last Packager: piernov
Votes: 4
Popularity: 0.55
First Submitted: 2025-10-18 10:00 (UTC)
Last Updated: 2025-11-03 17:45 (UTC)

Dependencies (5)

Required by (495)

Sources (35)

Latest Comments

1 2 Next › Last »

piernov commented on 2026-04-27 11:58 (UTC)

@verod9 as said below, the headers provided by this package are not made to be compatible with newer versions of GCC. Since software relying on CUDA is supposed to be built against a specific version of the CUDA SDK, if you want to install another package that depends on this version of CUDA (e.g., python-pytorch-cuda12.9), it is very likely that it will fail to build if you do not use GCC 14 or older. You can install this package by itself without gcc14, but it is most likely going to be useless. I would welcome patches to make it compatible with GCC 15, but I guess it is a lot of work.

Also, for PyTorch specifically, python-pytorch-cuda12.9 is even worse to get to build than gcc14, so I would instead recommend to use the upstream prebuilt Python wheels for CUDA 12.6 if you do not have a very powerful workstation and a lot of time.

verod9 commented on 2026-04-27 01:04 (UTC) (edited on 2026-04-27 03:08 (UTC) by verod9)

This package took extremely long to install (12+ hours) because of the gcc14 dependency. Is there a way to not include this dependency?

penguinx42 commented on 2026-03-15 23:56 (UTC)

I'm getting the same error as Skalli84

/var/tmp/pamac-build-user/cuda-12.9/PKGBUILD: line 102: cd: /var/tmp/pamac-build-user/cuda-12.9/src/builds: No such file or directory

piernov commented on 2026-01-23 16:30 (UTC)

@brannonking this package doesn't compile anything, it just ships libraries and headers that other software can compile with. It is when you try to compile something that depends on CUDA with an unsupported compiler that you can run into issues.

brannonking commented on 2026-01-23 15:51 (UTC)

This package doesn't need the gcc14 dependency. At least mine compiled without it just fine. Let's drop that (nontrivial) dependency.

brannonking commented on 2026-01-23 15:50 (UTC)

I too was confused by the call to install opencl-nvidia-590, so I cancelled out and installed the opencl-nvidia-580xx manually. I was doubly confused by that since I already had nvidia-580xx-dkms installed and the opencl-nvidia-580xx also seemed to unpack the nvidia binary driver. Dunno why I needed both of those (and maybe I didn't).

piernov commented on 2026-01-01 17:59 (UTC)

@tmoore multiple packages from AUR provide opencl-nvidia, such as opencl-nvidia-580xx, as well packages for older versions, so it is up to the user to choose the driver version they want. Even if it makes little sense to use it like that, it should still work with 590 from the official repositories as well.

tmoore commented on 2025-12-29 17:31 (UTC)

FYI. Don't know how this will impact this package using the opencl-nvidia 590 libraries, since 580.119 is the last one supporting pascal, but yours has a dependency on 590

opencl-nvidia-590.48.01-1

Skalli84 commented on 2025-12-13 16:55 (UTC)

Build fails for me. I'm on a 1080TI and cuda 13.0 is not working properly anymore, so I wanted to switch to 12.9.


==> Starting package()...
/var/tmp/pamac-build-user/cuda-12.9/PKGBUILD: line 102: cd: /var/tmp/pamac-build-user/cuda-12.9/src/builds: No such file or directory
==> ERROR: A failure occurred in package().
    Aborting...

I've replaced the username with user in the output. Before that the nvidia xterm ("NVIDIA CUDA PACKAGE") pops up saying "Unknown option: --target"

ofo commented on 2025-11-22 21:26 (UTC) 

  # Define compilers for CUDA to use.
  # This allows us to use older versions of GCC if we have to.
  ln -s /usr/bin/gcc "${pkgdir}/opt/cuda/bin/gcc"
  ln -s /usr/bin/g++ "${pkgdir}/opt/cuda/bin/g++"

Since this depends on gcc14, would these two lines need to link to /usr/bin/gcc-14 and /usr/bin/g++-14 respectively? (Default gcc is gcc-15 currently)

1 2 Next › Last »