Fedora 40 was released in April – same time as Ubuntu 24.04. When are we expecting support with the CUDA toolkit?
2 Likes
+1024
In the same environment, you need to use DNF to install CUDA Toolkit, CuDNN, and Tensor on Fedora 40 Linux. Currently, it only supports up to Ubuntu 24.04.
I don’t know if I can use DNF for this without specifying the RPM repo. And Nvidia has not published a repo for Fedora 40 packages. The installation instructions here show only Fedora 39: CUDA Toolkit 12.6 Update 2 Downloads | NVIDIA Developer
Just for the record. Fedora 39 is reaching end-of-life next month.
1 Like
This may not help many people for various reasons, but in my case I just use a makefile to build my fairly simple projects so …
.
F39 supports gcc 13, F40 gcc14, so if you upgrade to 40 it won’t compile, at the very least. However F39 clang is 17, F40 is 18, and cuda 12.6 update 2 supports clang 16 to 18 inclusive. So in my makefile I changed:
/usr/local/cuda/bin/nvcc -ccbin /usr/bin/gcc …
to:
/usr/local/cuda/bin/nvcc -ccbin /usr/bin/clang++ …
I already had clang but if not install it, and confirmed my projects were ok with clang (should be as its a supported compiler).
Upgraded to F40 without error, built projects, tested and profiled ok, and everything I use seems ok, and at least you could say the compiler is supported! But I can’t guarantee anything for anyone else.
I am aware there there are more civilised ways of specifying a different compiler or version of gcc but this is ok for my situation, and may suit some people with very simple builds.
alternatively upgrade to F40 but stay with gcc 14 and you will get this error message:
"error – unsupported GNU version! gcc versions later than 13 are not supported! The nvcc flag ‘-allow-unsupported-compiler’ can be used to override this version check; however, using an unsupported host compiler may cause compilation failure or incorrect run time execution. Use at your own risk. "
That will make that error go way but as it says: “use at your own risk.”
You have now swapped the EOL problem with F39 for having an unsupported rest of cuda;…
You will be in an unsupported state until there is a cuda that supports F40, then you could carry on with cclang or swap back to gcc.
Up to you to balance the issues or swap to ubuntu…
2 Likes
The latest Fedora CUDA packages are still for F39, which is EOL in just a few weeks. This isn’t leaving much time to continue to use CUDA on a supported Fedora system.
How can we get confirmation that Fedora is still supported by CUDA releases, or if not, can we confirm this now in order to give people time to deal with it?
Its a shame that this is still an issue.
Im also on Fedora 40 and am considering upgrading to Fedora 41 as soon as it comes out but seeing how I am still waiting for official support for gcc 14 and Fedora 40, I might just have to not upgrade whatsoever.
Just noticed that since a few weeks there is a cuda-fedora40 in Index of /compute/cuda/repos/fedora40/x86_64