This document discusses cache hierarchies and techniques to improve cache performance. It covers types of cache misses like compulsory, capacity and conflict misses. Methods to reduce miss rates include using larger block sizes, caches and associativity. Victim caches and prefetching can help tolerate miss penalties. The document provides examples of Intel cache designs including a dual-core chip with private 12MB L3 caches and an 80-core prototype with an entire die of stacked SRAM cache. Non-uniform cache architectures are discussed as an approach for large multi-megabyte caches.