1.The differences between CPU and GPU
   Design philosophy: CPU is designed to optimize for sequential code performance. The sophisticated control logic, and more importantly, large cache memory are its soul. However, GPU is designed to optimize for the execution of massive number of threads. Minimizing the control logic and cache memory are ways to make it efficient.

   Let's take a quick look at CPU's and GPU's architecture:

   GPU takes advantage over CPU in terms of memory bandwidth.
2.What concerns when we choose the processors for running our applications:
   The performance
   Market presence
   Easy accessibility
   Standard support(i.e.IEEE)
3.Know something about architecture of a CUDA-capable GPU

   Look at the pic:

   Host->Input Assembler->Thread Execution Manager->Building Block->Highly Threaded Streaming      Multiprocessors->Streaming processor
   (if any question, please make comments below!)
   One streaming processor usually contains a multiply-add(MAD)unit and an additional multiply(MUL)unit. Special function units perform floating point functions such as SQRT and RCP(reciprocal) SQRT as well as transcendental functions.
4.Trick for high speedup
   In general, applications often saturate the memory(DRAM) bandwidth, resulting in about 10X speedup rather. The trick is to figure out how to get around memory bandwidth limitations, which involved doing one of many transformations to utilize specialized GPU on-chip memories to drastically reduce the number of accesses to the DRAM.
5.Focus on computational thinking techniques. To see what is computational thinking, click here!