I've been trying for some time to install the CUDA toolchain (6.5) on my linux (Fedora 20 x64, GeForce GT 540M) system. The main issue is that the drivers that curently allow X to run (bumblebee-nvidia.x86_64) conflict with the real NVIDIA drivers (xorg-x11-drv-nvidia-libs-340.29-2.fc20.x86_64, and xorg-x11-drv-nvidia-devel-340.29-2.fc20.x86_64) that allow CUDA to run.
I had a wild party tossing bumblebee and installing CUDA only to find that on reboot X wouldn't start, and then reversing the whole process.
A friend told me that he had difficulty with the Nouveau ("Nouveau sucks" was the quote) which appears related to my bumblebee drivers (his instructions didn't work for me).
How does one go about installing the CUDA toolchain and drivers so that X starts? Sureley there's a way to get the xorg-nvidia drivers to do what bumblebee currently does?
As an additional potential solution: is it possible to have the X system run on my integrated intel graphics chip, and then use the NVidia card as compute only? How does one go about saying "Hey linux, use THIS chip with THESE drivers when running X"? That would be a first step. Then somehow having those drivers coexist with the nvidia drivers and then making the nvidia drivers compute only. If anyone could point to a good explination of how the whole X and xorg thing works, I would be highly appreciative.