Uncached

Uncached is a term used in computing to describe data, memory regions, or resources that are not stored in caches or are accessed without using caches. The concept is important for ensuring data integrity and predictable timing, particularly in contexts involving hardware devices or shared memory.

In CPU and memory systems, uncached memory regions bypass the processor's data and instruction caches. Accesses to such regions go directly to main memory or to the device hardware, which can result in higher latency and reduced throughput but avoids stale data and unintended reordering that caches might introduce. Uncached memory is commonly used for memory-mapped I/O, device registers, and buffers involved in direct memory access (DMA).

Operating systems and processors provide mechanisms to mark pages or regions as uncached, often using page table attributes or dedicated memory types. This ensures that reads and writes reach the actual hardware or memory without being cached, preserving correctness for certain types of operations and data.

In web and network contexts, "uncached" refers to resources or responses not stored by caches. Headers such as Cache-Control: no-store or no-cache, or directives to bypass caches, cause a client or intermediary along the path to fetch fresh content from the origin on each request, ensuring up-to-date information.

The main tradeoff of uncached access is a loss of performance due to bypassing caches. It is used when data freshness, strict ordering, or direct hardware interaction outweigh the benefits of caching.