I’d like to run CUDA programs on a GTX 1060 while having another graphics card control the graphics. My motherboard has Intel Integrated Graphics, and I initially wanted to have the Intel chipset run graphics with CUDA on the GTX 1060. However, after installing the nvidia driver I was encountering the login loop and was unable to figure out a way to get the nvidia card and the Intel graphics to play nicely.
Next, I tried installing a GT 218 (GeForce 210), however, the graphics driver that is compatible with the 1060 is not compatible with the GT 218.
I just want to be able to run CUDA on my GTX 1060 with a functioning X windows system. What is the best approach here?
You should be able to get the intel graphics working. Instructions on avoiding the login loop problem are covered (repeatedly) in other questions here on this forum.
The process to use the GT 218 would be similar. Since that GPU is no longer supported by current drivers, you would load the driver that supports the 1060, while adding the command line switch to not install the OpenGL libs, which are messing up the X stack that would work on the GT 218 (or intel gfx).
If you use the above method, then either the intel gfx or the GT 218 (which would be using the default OS driver, probably the nouveau driver) would have its graphics stack untouched, and should still work the same way after installation of the driver for the 1060.
Thanks! Once I used the “–no-opengl-files” command ( “sudo sh NVIDIA-Linux-x86_64-367.35.run --no-opengl-files”), it successfully installed the driver without causing the endless login loop.
I had read about the OpenGL drivers issue, but had been under the mistaken impression that it was only a problem when installing CUDA using the run file, and not also a problem when installing the driver using a run file.