I am using a cluster consists of both CPU based node and GPU enabled node. I have a regular account, and can log into a head node. CUDA toolkit and CUDA SDK have been installed in the head node. When I compile a .cu code that calls CUDA sub-functions, like cutCreateTimer(), it reports “cannot find -lcuda”. I know this library file may come with the CUDA driver, but for a cluster, the head node does not have a real GPU card, the problem is how can I install the CUDA driver on the head node without GPU hardware, or how can I obtain the library files coming with the driver without installing the CUDA driver. Thanks!
Please see my post in this forum, we have similar problems; it might be useful for you.