If you're writing CUDA code for GPU acceleration, you should really validate your hardware's device properties and memory management capabilities first (even if you have a fancy new card).