CUDA 3.2 deviceQuery works but Graphics fails Error while displaying graphical output

My GPU workstation has 2 x C-1060

Motherboard is Intel Workstation S5520SC http://www.intel.com/Products/Workstation/Motherboards/S5520SC/S5520SC-specifications.htm

Graphics controller is a NON-NVDIA integrated Server Engine* LLC Pilot II* Controller with 64 MB DDR2 memory, 8 MB allocated to graphics

We installed Ubuntu Server 10.10 64 bit, got all updates and installed CUDA 3.2.16.

Did my configuration as standard and compiled all the codes with the SDk.

Works fine. DeviceQuery output and all other text based outputs work smooth.

Issue is when i run OceanFFT or nBody or any other graphical output based codes, none execute !!!

No graphics display is shown. … Instead i get teh following error…

[font=“Consolas”]root@connoi-server:~/NVIDIA_GPU_Computing_SDK/C/bin/linux/release# ./oceanFFT [CUDA FFT Ocean Simulation][/font]

[font=“Consolas”] [/font]

[font=“Consolas”]Left mouse button - rotate[/font]

[font=“Consolas”]Middle mouse button - pan[/font]

[font=“Consolas”]Left + middle mouse button - zoom[/font]

[font=“Consolas”]‘w’ key - toggle wireframe[/font]

[font=“Consolas”][CUDA FFT Ocean Simulation][/font]

[font=“Consolas”]Xlib: extension “GLX” missing on display “:0.0”.[/font]

[font=“Consolas”]freeglut (./oceanFFT): OpenGL GLX extension not supported by display ‘:0.0’[/font]

[font=“Consolas”]root@connoi-server:~/NVIDIA_GPU_Computing_SDK/C/bin/linux/release#

Can somebody suggest What might be the issue here…Any help is much appreciated.

[/font]

You have the NVIDIA glx extensions installed, but no NVIDIA card. I very much doubt there is anything you will be able to do to get them to work. Even if you manage to link against a suitable GLX library (mesa3D in this case, I guess), those SDK examples rely on the CUDA-OpenGL interop facility, so that the CUDA context and X11 share a mapped OpenGL vbo in the same GPU memory. That probably won’t work with the display and compute GPU setup you are using.

Hi AVIDDAY,

Thanx for your reply. I am surprised, becuase Nvidia goes about saying that when using CUDA in Linux, we dont need to have a Nvdia graphics card installed simply because Linux already has support for OpenCL. When using with Windows a Nvidia card is a must. Any suggestions.

Regards

jsvarma

And it is absolutely true. If you have a CUDA compute device like a Telsa card, it is all you need to use CUDA. The examples codes you are asking/complaining about rely on NVIDIA’s OpenGL interoperability API and that requires an NVIDIA GPU with the NIVIDIA OpenGL driver. You could probably modify these examples to render frames offscreen and push the final pixel buffers to the onboard display GPU you have. It will be slow and ugly, but to coin a phrase “there is no such thing as a free lunch”.

Sorry but that makes absolutely no sense.

But that is because of the Windows driver design, it has nothing to do with CUDA, per se.

Caveat Emptor.