# Cache

**[Cache](https://en.wikipedia.org/wiki/CPU_cache)** is fast on-chip memory between the CPU and main memory, holding recently used data. Caches exploit temporal locality (reuse) and spatial locality (adjacent access), organized in hierarchy: L1 (~1KB, ~1ns), L2 (~256KB, ~7ns), L3 (~20MB, ~100ns), memory (~100ns+). Cache lines (64 bytes) are the unit of transfer.

[[cache-coherence|Multicore coherence]] adds complexity; [[false-sharing]] occurs when unrelated variables share a line.

## Example

Cache hierarchy on typical x86-64:

```
L1:  32 KB per core    ~1 ns    8-way associative
L2: 256 KB per core    ~7 ns    8-way associative  
L3:  20 MB shared    ~100 ns   20-way associative
RAM:  —              ~100 ns+
```