Cuda support for OpenSuSE LEAP 42.2?

Does anyone know how/if Cuda 8.0 can be installed on OpenSuSE LEAP 42.2? I am aware its “not supported”, but its unclear if it can be make to work (with reasonable effort) or not. Its generally not feasible to change the default GNU compilers with OpenSuSE.

If not, can someone at nVidia address this? (eg, will OpenSuSE support continue? if so, when?)

tnx … John G. Shaw

By definition, “not supported” = you are on your own, don’t expect help from NVIDIA. As for future plans, NVIDIA seems to have a firm policy of never publicly commenting on unreleased products, and have actually stated that explicitly a few times in these forums.

Checking the Linux Installation Guide for CUDA 8.0, it shows support for OpenSUSE 13.2 (kernel 3.16.6, gcc 4.8.3), why not use that?

Because 13.2 is at end-of-line and will be unsupported as of january.

Many others must have tried leap versions with cuda, but I only found a couple of posts about it. I was hoping someone who got this to work, or concluded it cannot, would respond.

It would be usful for nVidia to publish the underlying requirements for a release such as: gnu (or other compiler) version number + related libc version (eg, gcc 4.8.x), kernel version (3.x), etc. The largest supported gcc version looks to be 5.x, and leap 42.2 uses 6. This does not mean it won’t work (eg, comment out the check in the appropriate cuda header file). Most supported distributions are back on 4.x.

Correct me if I am wrong: It seems that OpenSuse 13.2 came out in November of 2014, meaning it’s just over two years old at this point. Why would it be EOLed after such a short time? I am puzzled.

My understanding is that NVIDIA tests particular Linux distros with their corresponding default gcc, glibc, and kernel versions, and these component versions are stated in the Installation Guide.

If you choose different components to go along with a given supported distro, you are on your own. From past reports in these forums, cobbling together components and removing the safety checks from the CUDA header files sometimes works, or it can result in all kind of cryptic errors.

From a practical perspective, I don’t see any realistic alternative to NVIDIA’s validation process. They are not in the business of figuring out what Linux components may or may not work together harmoniously, instead they rely on the distro configurators to figure that out, then just test against a specific distro.

Which distros are supported by CUDA is presumably driven purely by customer demand, so maybe OpenSuse is no longer as popular as it used to be among people using CUDA on Linux? Presumably NVIDIA can gauge demand easily by counting the number of downloads for a given distro’s package.

IMHO, some consolidation in the Linux space wouldn’t hurt.

Agreed to everything you said. SuSE’s stated support is about 2 years past their next release. EOL for 13.2 is stated as January 16, 2017 (Lifetime - openSUSE Wiki).

Its next to impossible to switch the GNU compilers in the distros, even if they are in the repositories (much grief can ensue). I would not mind fiddling with cuda headers if its a temporary fix, and I know a supported version is coming “soon”.

I program cuda applications (complex physics codes) and its painful to loose a stable working distribution so soon (crap). I like OpenSuSE for home use as YaST makes it very easy to maintain (at least easier than the Redhat variations). I use RHEL 6.x and 7.x at work, but maintenance is someone else’s problem. I may end up having to give up on SuSE if nVidia chooses not to support it anymore. So far, the only “support path” that I would fully trust seems to be RHEL (CentOS). (Windows & MacOS does not count;)

The one thing I can say about RHEL (which I used for many years), is that I never encountered a problem with it, and I recall at most a handful of problems related to CUDA on RHEL (or CentOS) being reported in these forums (as opposed to the myriad issues reported with Ubuntu, which I chalk up to the OS itself, not CUDA). So from that perspective, I can definitely recommend RHEL.

My old home computer finally failed, and I ended up replacing most of it (Skylake CPU and z170 chipset + memory). This was the reason for my original post. I installed OpenSuSE LEAP 42.2 and the latest nVidia drivers (375.26) using the repository method. I then extracted the Cuda toolkit from the Cuda “run file” installer.

It all works with out any issues. I compiled all the samples, ran several of them (eg, fluidGL), and finally compiled and ran my own Cuda codes. I have to conclude that it “just works”. Perhaps the developers can make it official.