Traditionally Data mining involves analyzing and moving alot of data around. In this context what is the major bottlenecks in the current NVIDIA CUDA archiecture for this type of application.?