A benchmark is a point of reference by which something can be measured. In surveying, a "bench mark" (two words) is a post or other permanent mark established at a known elevation that is used as the basis for measuring the elevation of other topographical points.
In computer and Internet technology, the term may have any of these meanings:
1) A set of conditions against which a product or system is measured. PC magazine laboratories frequently test and compare several new computers or computer devices against the same set of application programs, user interactions, and contextual situations. The total context against which all products are measured and compared is referred to as the benchmark.
2) A program that is specially designed to provide measurements for a particular operating system or application.
3) A known product with which users are familiar or accustomed to that other newer products can be compared to.
4) A set of performance criteria which a product is expected to meet.
Laboratory benchmarks sometimes fail to reflect real-world product use. For this reason, Eric Raymond defines a benchmark as "an inaccurate measure of computer performance" and cites the "old hacker's saying" that "In the computer industry, there are three kinds of lies: lies, damn lies, and benchmarks."
Still, benchmarks can be useful and some companies offer benchmark programs for downloading or a benchmark testing service on their own site.