Right, so I've seen that a few people, instead of having a "real" matrix (as in, an array of arrays) use a single array, and when they want to access position (x, y) do something along the lines of array[x + y*width].
Is it really worth it? You're trying to trade two memory accesses for a single one with a little arithmetic on the side, right? Is that speed gain even significant?
Is there something I'm missing?