The Compute Capabilities designate different architectures. In general, newer architectures run both CUDA programs and graphics faster than previous architectures. Note, though, that a high end card in a previous generation may be faster than a lower end card in the generation after.
From the CUDA C Programming Guide (v6.0):
