Beware of changing graphics drivers in Linux …
I write this to relate my experience installing proprietary drivers, and updating graphics drivers or changing them, in Linux. Late in 2013, I installed NVIDIA’s CUDA package to use the GPU as form of “parallel processor” to be used with the CPU and an extension of the C language to enhance performance when computations benefit from parallelization. The test programs went well, achieving 9 GFlops on matrix computations. Subsequently, I did bitcoin mining with the graphics card or GPU at too intense a level, which caused the GPU’s fan to fall off. Not wanting to take risks, I bought a new graphics card (evga GeForce GT 640 dual-slot, 2 GB video memory), and installed it without problem. The updater with CentOS 6 series peridically suggests system updates, including new Linux kernels. After updating to a newer kernel, the X-server which allows “windowing” and graphical user interface wouldn’t start. This still allowed to used older kernels from the Grand Unified Booloader (or GRUB) menu, at start-up. Selecting the “next-to-latest” kernel in GRUB worked fine and the X-server worked fine. Yesterday, I installed the drivers provided by NVIDIA for my newer graphics card, thinking that I could then boot from the latest Linux kernel and that the X-server would then start and work properly. Unfortunaly, this made things worse and I was left with no functioning X-server “windowing” system …
After an hour or so of experimentation, in console mode, no GUI, I ran the old CUDA installing script, namely:
# sh cuda_5.5.22_linux_64.run [ENTER]
from the directory containing the script,
This install both NVIDIA graphics drivers and the CUDA package, the one allowing for use of a GPU in parallelized computations using an extension to C developped by NVIDIA. This fixed everything. I rebooted into the latest linux kernel in the GRUB menu, and the X-server started completely normally.