Quadro and Tesla

I have both a Quadro FX1700 and a Tesla K20 installed on the same linux (opensuse) machine. I installed the latest version of the driver (346.89) together with cuda-7.5. This works fine as long as I only run non-graphical codes from the cuda samples.

However, the driver does not handle the Quadro card. when I run graphical samples I get the following error:

Xlib:  extension "GLX" missing on display ":0".
freeglut (./marchingCubes): OpenGL GLX extension not supported by display ':0'

How can I install another driver specifically for the Quadro card without disabling the K20?

Thanks in advance.

Only one NVIDIA GPU driver can be in use at any given time on linux. Your Quadro FX1700 is a cc1.x device, and is not supported by CUDA 7, CUDA 7.5, or any driver that would support CUDA 7 or CUDA 7.5

The removal of support for cc1.x devices has been publicized in release notes and elsewhere.

If you want to support both devices, you must revert to an older driver that supports your cc1.x Quadro device.

The latest CUDA toolkit that supported it was CUDA 6.5, so that is one possible option. Install CUDA 6.5 and the 340.xx driver that comes with it, and both devices will be supported.

Thanks for your answer. It still gives the same error if I replace the quadro card by an ATI Radeon. Isn’t there any way to display the GPU-computed animations (cuda 7.5) with another non-cuda graphics card ?

The “GPU-computed animations” samples involve (on linux) cuda/opengl interop. AFAIK this interop requires NVIDIA GPUs, and NVIDIA OpenGL support.

So if I understand correctly in order to get cuda 7.5 working I have to get another recent enough gpu-enabled nvidia card, so that I can visualize what’s computed by the K20?

Yes, that should work.