It is generally believed that base scores are the better choice when comparing. Base scores require that the same set of compiler options be used for all tests. Peak scores allow a different set of compiler options be used for each individual test. Peak scores are much more finely tuned and may not be as representative.