Other than in a prior post I found it rather easy to install the driver under Fedora 12 and I even got CUDA working. NVIDIA’s Linux driver installs without any problem and the CUDA Toolkit also installs smoothly. The only issues you have to face are errors of the type
... inline function ... cannot be declared weak
Just comment out those lines in the corresponding header files. I am not sure if and to which error it leads, but until now I could not find anything abnormal in my results. Getting the SDK to work is a much more difficult job, and according to entries in NVIDIA’s Linux forum requires quite a lot of work-arounds. Fortunately I don’t rely on any of the stuff contained in the SDK, and since the FFT libraries, which are the most crucial part of my code, work as expected I am satisfied with that solution. BTW, the CUDA Profiler for Linux has been improved tremendously and offers now really nice functions which help you to get the critical parts of your code tuned to squeeze out maximum performance of your GPU.

Popularity: 37%