Problem 1: True, False, or just Explain what you think. (15 pts)

• The dreaded true-false section has magically appeared. Three points each, and be a tad careful, as some are easy, and some are tricky. Circle the number of each False statement. If you prefer, you may just explain why you think something is false for the same credit.

1. The number of instructions in a program and the number of clock cycles required for each instruction are independent.

   Answer: False.

   Why? because number of instructions is IC, and the number of clock cycles per instruction is CPI, and BOTH depend on the ISA of the machine on which they are being executed. Since they share a common source, and modifying the ISA could cause changes in both, it's hard to consider them as independent. Now, if you say that you are assuming a given architecture or ISA in your ``reason'', I'd think that partial credit would be available. However, in general, they aren't statistically independent.

2. The performance of a computer architecture is usually improved by decreasing the clock rate.

   Answer: False. Be careful here. Decreasing the clock rate means that there are FEWER ticks per second, or more clocks per tick, which is equivalent to making the clock slower. Remember, units of rate are cycles per second, and time are seconds per cycle.

   Also remember that performance is the inverse of clock cycle time because performance is a rate.

3. A Java program will execute faster on a machine with a 1.2GHz clock than on a machine with an 800MHz clock.

   Answer: Who Knows?

   This one is a bit nasty. If your Java virtual machine is purely software AND you have a small RAM, it may be that you can't improve the performance by making the clock faster because your process is memory-poor and will be stuck page faulting. That is, you don't have enough memory to be able to take advantage of the quicker clock.

   If you are using, say, a Java chip--and there are some, or you have enough memory, it is indeed likely, if all other parameters remain unchanged, you will see improvement in the execution time.

4. Modifying the ISA in a way that decreases the average clocks per instruction has little or no effect on the average number of instructions per program.

   Answer: False. This is NOT sneaky. It's the same idea as in the first problem in this section. When you modify the ISA, both the IC and the CPI may change because the program still has to do the same job, but using a slightly different implementation for some of the instructions, and even eliminating or adding instructions so as to change the average CPI. Thus, it is naive to expect no change in the IC when the CPI changes.

5. The MFLOPs rating of a given computer architecture is a constant. Answer: False Remember, MFLOPS is millions of floating point instructions per second. Since different programs will use different instructions that take different amounts of time to execute, there is no reason to expect the MFLOPS rating to be constant for two programs. Some programs may use floating point heavily; some may never invoke an FP instruction. Thus, there is no way to assure that the same MFLOPS rating will be computed for a specific architecture because MFLOPS depends on the program being executed.

Web Accessibility