For each configuration tested, I present the results within the context of "all WinMarks." The results show charts of the entire series of "all WinMarks. The entire series is presented for several reasons. First, possible interactions between subsystems might occur. Testing of two peripheral devices might yield similar performance in the subsystem of interest, but affect the performance of other subsystems. Second, benchmarks are reported only if an entire series of "all WinMarks" consistently completes. This is my criterion for stability. You can tweak a system to optimize a particular mark. The result is a high score, but the system might be unstable and, thus, not usable in real world computing. As alluded to above, tweaking one subsystem commonly decrements performance elsewhere. Third, inspecting other subsystems verifies that overall system performance is comparable. This strengthens conclusions regarding the particular configuration parameter in question.
So what's the problem with WinBench 98?. I know that I am an honest person and resist temptation.:) This, you don't know. The charted data from Winbench 97 is derived directly from a binary coded database. I don't know how to modify the data even if I wanted to. An Excel chart can be modified with the simple edit of a cell. This simplicity is a potential abuse of the benchmark results. When using the native charting of Winbench 97, you either chart it as is or you don't chart it at all.