This question is answered very clearly by
I am just adding a short reference that has a two sided use,
- Gives you a bit more breadth in the Cache theory
- useful for understanding newer architectures (Nehalem...)
- For all those memory performance power users on this site,
- gives you an idea of how caches work and which sizes matter when deciding your PC
Memory part 2: CPU caches at the LWN.net site (Oct, 2007).
Editor's note: This is the second installment in Ulrich Drepper's "What every programmer should know about memory" document. Those who have not read the first part will likely want to start there. This is good stuff, and we once again thank Ulrich for allowing us to publish it.
The longish article may also help understand why the caches were moved into the processor module (as against the old days described by
cwrea in the comment above, which are better forgotten).
Nehalem L3 cache note at ExtremeTech.
An old Overclocking article reference that I did not include earlier specifically because it does not apply to L2 Cache scaling. It is interesting to read in the context of my comments to another answer here (by
From Three Gems for an Overclocker: on the Intel Celeron 2GHz,
Intel Celeron were always based on the same cores as the faster processor families, with that only difference that the L2 cache was twice as small, bus frequency was reduced and the clock frequencies were lower. As for the cache, there is no way to get its cut-down half back, however, in terms of frequencies, overclocking comes to rescue and allows speeding up the low-cost processors immensely. Not so long ago, following in the footsteps of Pentium 4, Celeron processor family acquired a 0.13micron Northwood core. The first Celeron CPUs based on it appeared Celeron 2.0GHz. As we have expected, they appeared very easy to overclock. Their core frequency can be raised up to that of the fastest Pentium 4 models, which is roughly 3GHz. And only the cut down 128KB L2 cache, prevents Celeron from beating all overclocking records.