How to use device level API in CUDA 6?

I’m having link errors whenever I use the device level functions (e.g. cuInit()). I tried searching for that symbol in the libs and see some of those functions in libcudart_static.a, but with a version prefix (so that one SDK can compile for multiple runtime versions?).

How can I use those functions? Thanks.

False alarm. Apparently, libcuda.so isn’t part of the SDK, but the driver instead and I was compiling on a machine that didn’t have a CUDA driver installed (I’m using the Intel compiler and only have a license on a few machines).