CUDA on Non GDM platform


I work in a product based company and I want to use CUDA on our own customized platform which uses a stripped down version of Red Hat Linux 5.x. We don’t have gdm installed on it as we don’t need a UI for this system. Can I install nvidia driver and CUDA on this platform without installing gdm or xserver? Do I need libraries like openGL etc also?


You can install CUDA on a machine without X11. The installer will run and the drivers will be installed correctly. Without X11, you will need to create the /dev/nvidia? and /dev/nvidiactl

devices (this is discussed in the installation instructions), but it will work. I have headless cluster nodes running this way with Centos 5. AFAIK, you don’t actually need openGL installed unless you are planning to use it, although I haven’t tried it.

You do not need OpenGL installed to run CUDA on a headless node. The problem you will run into is that if you try to build the codes in the SDK, many examples use OpenGL. You will have to modify the Makefiles to skip those codes so you can build the rest.