9) You have access to CPU's capable of performing at 1 MIP each. Your goal is to design a new system that has 2 MIPs of performance. You can either redesign 1 of the current processors such that it is twice as fast, or use two of the existing processors in a parallel processing fashion. Which approach would generally give a better cost/performance ratio? Use a simple diagram to defend your answer. [3 pts] 10) One of the design goals of Instruction-Level Parallelism (ILP) was to make it transparent to the software layer; however, there are limitations as to how effective it is, and to what degree of aggressiveness or "look-ahead" it can be performed at. What are the causes of the limitations and how do they arise? [3 pts] 11) What is HyperThreading, and what, if anything, should software programmers do to leverage it as effectively as possible. 3 pts] 12) What are the micro-architecture and macro-architecture trends and HW design problems that have led to the need for programmers to explicitly express parallelism into their software? What are some of the challenges associated with directly exposing the underlying hardware parallelism to the software layer and programmers ? [3 pts] 13) Describe the four Flynn Architecture Classifications. Explain why the MISD configuration has traditionally thought of as non-sensible. [3 pts] 14) In some ways, the speedup from pipelining is "an illusion" because each instruction may actually take longer to complete in a pipelined machine as compared to a non-pipelined one. Explain how and why this statement is true. [3 pts] b) Under extremely poor conditions, the pipeline could conceivably be stalled and flushed after every single instruction. In this situation, how would the performance of the pipelined machine compare against that of the non-pipelined one? Why? [2 pts]