CUDA Support for Linux distros that aren't end of life

CUDA 8.0 was released September 28, 2016. At that time, Fedora 24 had been out for three months, but CUDA 8.0 only officially supports Fedora 23 which went end of life December 20, 2016, meaning it will no longer receive any security patches. I work at a University, and the IT department requires that you be running an up-to-date operating system, so I have to make sure that I have a currently supported version of a Linux distro installed.

Now, I have been able to get CUDA working on Fedora 25, after making some changes to a header file, but I have to compile using --compiler-options “-std=c++98” to avoid problems, meaning I can’t use any of the more modern features of C++ in my files containing device code.

Of course, one option would be to use one of the Ubuntu LTS versions, but for several reasons I don’t like Ubuntu, and I like that Fedora tends to incorporate newer versions of compilers and other software very soon after stable releases.

My question is, does anyone know if the CUDA devs plan on releasing updates that incorporate support for newer versions of some supported Linux distros? The only thing that seems to hold back support for the newest version of Fedora is that Fedora 25 uses GCC v6.3.1, and CUDA 8 only seems to support up to v5.3.1. I do know that there are compatibility issues (hence needing to use -std=c++98 for compilation because of the gcc v6+ implementation of the c++11 and later standards), but it seems like these could be addressed.

Not sure about Fedora, but in Ubuntu you can easily install a previous GCC version of the compiler and set it as default with a simple update-alternatives command. I can only assume they offer a similar method in their distribution.

As for your update question… no, they do not release updates on the same version. Depending on when the newer O/S releases and as they qualify them as ‘compatible’ they will issue the release notes mentioning the supported configuration for the new versions. /thread

Thanks for the reply, vacaloca! I was afraid that would be the answer.

I have setup previous versions of GCC before, when I first started using CUDA. For the past couple of versions (since 5 I think), I’ve just made an edit to host_config.h so that it doesn’t flag an unsupported version of GCC and things have worked. This still works for CUDA 8, just have to make sure to specify -std=c++98 for GCC since the newer versions default to gnu++14, and that causes the compatibility issues.

It would just be nice to not have to resort to these measures to have CUDA working.

Thanks again for the info!