Discrete GPU makes sense in a premium model, but for the base model cost savings an APU grants are of the essence. SmartShift in an APU can work smoother and faster than in a discrete CPU and GPU solution. I can see AMD adding in premium models a discrete GPU of the same model and speed of the one in the APU, as even if parallelism can be done with different units, using the same at the same speed gives the maximum efficiency and minimises possible problems.