Driven by the curiosity for GNOME 3.0 and a little bit fed up with the Ubuntu installation on my GPU laptop I decided to install F15 on that machine. All my other PC’s are running F12 or F13, which are either supported by CUDA. As there is no official support for F15 at the moment I was a little bit suspicious if I get the video driver up and running and , more important, if the CUDA compiler does what it is expected to do. Based on lessons from the past I did the following steps:
1.) Install F15 with the “Install with basic video driver”. This is crucial as CUDA seems to have a hard time working with the nouveau driver (actually I have no idea if they work together at all)
2.) After installation finished, I downloaded the latest video driver (270.41.19) for Linux 64bit, did a “/sbin/init 3″ and installed the driver without troubles
3.) Rebooting the system and GNOME 3.0 comes up. Thus I assume the driver is doing the correct job.
4.) Installation of the CUDA toolkit went without any troubles.
5.) Since F15 comes with GCC 4.6 the header file /usr/local/cuda/include/host_config.h needs to be modified (line 80) in order to remove the error of the version mismatch. I just replaced the last “4″ with a “7″
6.) I could compile my GPU code without any troubles and until now it seems to work as expected.
7.) Additionally, I tried to get the SDK examples running, but compilation failed after the 2nd demo code due to some troubles with threading. Anyway, two examples (“oceanFFT” and “simpleGL”) were compiled successfully and they work as expected.
So in summary I would say that CUDA should run under F15 (at least for my code). The hurdles during installation are not very high, and hopefully disappear totally in future.