I have been running the Nvidia proprietary driver for a few days without any problem. However, this afternoon I have installed CUDA SDK and CUDA Toolkit and run into the following issues:
1.- It seems that the latest CUDA SDK requires gcc-4.7 (4.6 installed) and therefore a previous version was installed.
2.- CUDA samples compile fine. Samples not requiring X also run fine, however those requiring X fail with "CUDA-capable devices unavailable or busy".
3.- Booting hangs for a while after loading the Synaptic module then it continues.
4.- Most of the times X does not start. I cannot go into a console to know what it is going on, but sometimes it does start.
My guess is that maybe one of the CUDA packages (I have also installed PyCUDA) has installed its own drivers and now I have two drivers in the system.
Is this possible? Should I remove the previous driver?
The problem seems solved, but I will test it a bit more before marking the thread as solved. This is what I did:
1.- I installed gcc-4.7 from the Sekh overlay in order to be able to install version 5.5 of the CUDA Toolkit (with gcc-4.6 the version installed was 5.0.35, both are in the Portage tree).
2.- I removed the beta driver merged from Portage and replaced it with the latest certified driver from the Nvidia site.