[ethbr0]

I think of it in terms of computer memory levels.

There are some computations (synthesis) that require so many (non-front loadable) memory accesses that it's impractical to do them from memory with significant delay (books), because the number_of_accesses * access_time dominates the project time.

Instead, you must have a working set of core information (or at least pointers to information) in low-latency memory (your brain).

Example: How much longer would it take me to do a multi-digit multiplication if I had to look up the process for multiplication in a book for every digit multiplication I did? And what if that multiplication were just one of many in the higher-level math problem I was trying to solve? (Then generalize to any problem that requires a core base of knowledge)

[kristianbrigman]

It is very similar to caching performance impacts. And like you say, sometimes performance is just faster, and sometimes it actually enables functionality…