History of Memory Cache on PCs
This section is only for those interested on the historic aspects of memory cache. If you are not interested on this subject you can simply jump to the next page.
Memory cache was first used on PCs at the 386DX timeframe. Even though the CPU itself didn’t have memory cache inside, its support circuitry – i.e., the chipset – had a memory cache controller. Thus the memory cache at this time was external to the CPU and thus was optional, i.e., the motherboard manufacturer could add it or not. If you had a motherboard without memory cache your PC would be far slower than a PC with this circuit. The amount of available memory cache varied as well depending on the motherboard model and typical values for that time were 64 KB and 128 KB. At this time the memory cache controller used an architecture known as “write-through,” where for write operations – i.e., when the CPU wants to store data in memory – the memory cache controller updates the RAM memory immediately.
With the 486DX processor Intel added a small amount (8 KB) of memory cache inside the CPU. This internal memory cache was called L1 (level 1) or “internal,” while the external memory cache was called L2 (level 2) or “external.” The amount and existence of the external memory cache depended on the motherboard model. Typical amounts for that time were 128 KB and 256 KB. Later 486 models added the “write back” cache architecture, which is used until today, where for write operations the RAM memory isn’t updated immediately, the CPU stores the data on the cache memory and the memory controller updates the RAM memory only when a cache miss occurs.
Then with the first Pentium processor Intel created two separated internal memory caches, one for instructions and another for data (at the time with 8 KB each). This architecture is still used to date, and that is why you sometimes see the L1 memory cache being referred as 64 KB + 64 KB, for example – this is because there are one 64 KB instruction L1 cache and one 64 KB data L1 cache. Of course we will explain later what is the difference between the two. At that time the L2 memory cache continued to be located on the motherboard, so its amount and existence depended on the motherboard model. Of course having a system without memory cache was insane. Typical amounts for that time were 256 KB and 512 KB.
On AMD side K5, K6 and K6-2 processors used this same architecture, with K6-III having a third memory cache (L3, level 3).
The problem with the L2 memory cache being external is that it is accessed with a lower clock rate, because since 486DX2 the CPU internal clock rate is different from the CPU external clock rate. While a Pentium-200 worked internally at 200 MHz, it accessed its L2 memory cache at 66 MHz, for example. Then with P6 architecture Intel moved the memory cache from the motherboard to inside the CPU – what allowed the CPU to access it with its internal clock rate –, except on Pentium II, where the memory cache was not located inside the CPU but on the same printed circuit board where the CPU was soldered to (this printed circuit board was located inside a cartridge), running at half the CPU internal clock rate, and on Celeron-266 and Celeron-300, which had no memory cache at all (and thus they were the worst-performing CPUs in PC history).
This same architecture is used until today: both L1 and L2 memory caches are located inside the CPU running at the CPU internal clock rate. So the amount of memory cache you can have on your system will depend on the CPU model you have; there is no way to increase the amount of memory cache without replacing the CPU.