Bug#1142932: nvidia-cuda-toolkit-gcc: depends on gcc-13

[Michael Prokop]

Hi,

* Dominique Belhachemi [Wed Aug 05, 2026 at 07:55:23PM -0400]:
> On Tue, 28 Jul 2026 12:54:16 +0200, Emilio Pozuelo Monfort wrote:

> > nvidia-cuda-toolkit depends on gcc-13, which won't be shipped in forky.
> > Please update to 15 or 16, or ideally use the default version (currently
> > GCC 15).

> CUDA's 12.4 nvcc refuses anything newer than gcc-13.  crt/host_config.h:143:
> 
>   #if __GNUC__ > 13
>   #error -- unsupported GNU version! gcc versions later than 13 are not
>   supported! ...
>   #endif
> 
> You could probably patch it out or overwrite it
> by defining __NV_NO_HOST_COMPILER_CHECK
> but it is not recommended by upstream.
> 
> The minimal valid configuration to fix the bug is probably driver 580
> and CUDA 12.9 (with gcc-14 support).
> To be safe, I would go with driver 580 and CUDA 13.0 (with gcc-15
> support). Pre-Turing GPUs still run on 580, but CUDA 13.0 dropped
> compilation support.
> But Forky should probably ship something newer (e.g. To better support
> local AIs etc ...)

The latest CUDA version by now is 13.4.2, and NVCC version being
13.4.92, released as of 2026-09-16, see
https://developer.download.nvidia.com/compute/cuda/redist/redistrib_13.4.2.json.

As stated by Dominique, CUDA 13.0's NVCC Compiler supports GCC 15
and Clang 20 (see e.g.
https://developer.nvidia.com/blog/whats-new-and-important-in-cuda-toolkit-13-0/).

The latest CUDA version 13.4 also supports GCC 16 and Clan 22,
according to
https://developer.nvidia.com/blog/cuda-toolkit-13-4-adds-windows-on-arm-support-and-greater-control-over-shared-gpus/

Would be great to have a current version of CUDA in forky. :)

regards
-mika-
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://alioth-lists.debian.net/pipermail/pkg-nvidia-devel/attachments/20260929/5b8c9a2a/attachment.sig>