Over on stackoverflow someone posted a link to a paper written by a Linux kernel engineer about how to use computers and RAM. He started off by explaining how RAM works (right down to the flip-flops) and then went on to discuss performance problems associated with operations on matrices (column vs row accesses), offered solutions and then dealt with some stuff MMX instructions can do.
Sorry it's a bit vague but I can't find it anywhere. I think the guy had a Scandinavian name, possibly Anders
