A summary of methods to measure the performance of a CPU

Take some physical measurement and use it to compare CPUs.
e.g. Clock rate, MIPS(millions of instructions per second)
Problem: such measurements don't corrrectly gauge performance.
Run an (important) real program
Measure cpu time; faster is better.
Problems: ports are expensive; it may be difficult to recreate the platform
Run a set of real programs (workload)
Determine which workload has the "best" throughput (total time for workload)
You cannot sum cpu times.
Problem: This is infeasible if you must port the workload.
Run fake programs designed to measure performance (benchmarks)
There is a major industry to market, write, port and run benchmarks.
Problem: It's easy (and common) for vendors to cheat.