Many believe that one gigabyte of RAM is enough for virtually any non-professional application scenario. Unfortunately, that is not the case. Have you ever launched a recent gaming title with lots of resource-hungry applications already launched? Everything might seem fine as Windows relocates the applications' memory data to the swap file on your hard drive. However, as soon as you hit a Windows key accidentally, the OS will hectically try to exchange the gaming data in the main memory by the application data that was swapped before
Cached Vs. Non-Cached Benchmarking
When using Doom 3 for benchmarking, you initially run the demo a few times and ignore the first results since they are lower. The reason for the lower result is that the game reads texture and map data from the hard drive to system memory, and hard drive access is a FPS killer.
If you don't shut down the game after the first run, the demo will be cached in the system memory, thus avoiding the hard drive access penalty the next time you run it. If you're testing a CPU or a graphics card, you obviously won't take the first run into account since you don't want the hard drive access to affect your results.
However, the testing methodology used to determine the RAM's impact on performance is different. In this case, gauging performance involves the benchmark results generated the first time you run the demo.
Think about it: Did you ever load a saved game, complete the level, load the same saved game again and play the level one more time just to avoid low performance caused by hard drive access? The resulting performance gain is why we included the first runs in the graph - and don't worry, the first run results were very consistent.