A detailed look at some popular benchmarks

‘Fair benchmarking’ can be less of an oxymoron — if the people using benchmark results know what tasks the benchmarks really perform and what they measure. This paper characterizes in detail the most commonly used industry standard benchmark programs (Whetstone, Dhrystone, Linpack) and names a number of pitfalls that users have to be aware of. Other benchmarks are mentioned, and the relative merits of small vs. large benchmarks are discussed. The often neglected non-CPU influences on benchmark results are discussed, and ‘consumer advice’ if offered.