My undergraduate algorithms course did point out that memory hierarchies and branch predictors laugh at big-O notation, and that you should always (as Roboprog

My undergraduate algorithms course did point out that memory hierarchies and branch predictors laugh at big-O notation, and that you should always (as Roboprog said) measure measure measure instead of just assuming a particular algo is best. It was one little piece of material in a course with a lot of ground to cover though, so it's hard to say how well it stuck with the class...