why L1 cache and L2 cache are different sizes

[jonkcarter]

Hello i just have a question really i want to know why L1 cache and L2 cache are different sizes if any one could help me with this question i would really like to hear from you thanks.

[jlreich]

I am not an expert on this but from my understanding the biggest issue, in a nutshell, is that a smaller cache is faster, but a larger cache will take less hits, meaning it is more likely to hold the sought after data. So to improve performance the CPU will go to the smaller faster L1 cache first, if the sought after data isn't there it will then move to the larger relatively slower L2 cache, then to the L3 (if present), then to main system memory, then finally to the HDD if necessary.

So basically there is a balance between smaller and faster, and larger and slower but taking less hits.

There is much more to this though. In fact you could probably write a masters thesis on cache alone.

[Sylvander]

COST
The faster the cache memory, the more it costs; the slower, the less it costs.
Hence it is a smart strategy to use small amounts of fast expensive memory at the top of the memory higherarchy [probably modelled on our social structure (very few millionaire's at the top and millions of poor grunts at the bottom)], and large amounts of low cost storage at the bottom.
The stuff that's used the most [most needed/valuable] moves up the higherarchy; the stuff used the least [least needed] goes down the higherarchy.

[jlreich]

Well yes cost is a consideration. If SRAM wasn't so expensive it would be used as main system ram instead of tiny amounts for cache. But the smaller and larger sizes as you move down the ladder has to do with branch prediction and performance. The smaller size is basically the same concept as a smaller registry; less space to look through equals less time to find what you are looking for, equals better performance, if the data is actually there. If not then move too the larger but slower (in ns) L2, but still way faster than regular SDRAM system memory. Now hopefully the CPU doesn't need to go all the way down the to the HDD to get data. That's like walking to the other side of the world to get a stick of butter.

[jlreich]

but a larger cache will take less hits, meaning it is more likely to hold the sought after data.
So basically there is a balance between smaller and faster, and larger and slower but taking less hits.

Sorry, that should have been misses. A hit would be actually finding the data.

[saphalline]

Part of it is the nature of the evolving architectures of computers and their processors. In the early days of computing, CPU's didn't have cache. It wasn't an idea that had developed until the late 1970's (at least for PC processors). There are many reasons for this, but performance has always been the driving factor behind caches.

As to the reason why L1 cache is bigger than L2 cache - the simple answer is that they aren't. In the world of x86 CPU's, Socket 7 designs were known to support but not require L2 cache. In later years, the Duron was a cost-effective "cheapo" CPU from AMD that had less L2 cache than L1 cache. Intel has never been able to do this due to the fact that all of their CPU's use an inclusive cache architecture, while AMD uses exclusive. Other processor platforms have similar stories, such as the SPARC series in Sun-based servers and the Motorola/PowerPC CPU's of the Apple Macintosh PC's.

I'm sure if I dig deep enough into CPU history, I could find many examples of CPU's with equal-sized L1 and L2 caches. So the "rules" as we see them aren't always absolute.