Essential Terms

A mathematical formula used to find the maximum theoretical improvement of an entire system when only a part of it is upgraded or parallelized.

A mechanism in multiprocessor systems that ensures all caches maintain a consistent view of shared memory. When one core modifies a cached memory location, coherence protocols update or invalidate stale copies in other caches so that future reads observe the correct value.

The CPU looks for a piece of data in the ultra-fast cache memory, finds it immediately, and reads it with minimal latency.

The CPU looks for data in the cache, doesn't find it, and is forced to drop down to slower main memory (RAM) to fetch it, creating a costly performance penalty.

The physical latency window or time lag required for a carry bit to travel and stabilize through a chain of adders. In a Ripple Carry Adder, Full Adder 4 cannot calculate its final Sum until Full Adders 1, 2, and 3 have sequentially completed their operations and passed the carry bit down the wire.

Occurs when the processor cannot determine the next instruction to fetch because the outcome of a branch, jump, or conditional instruction has not yet been resolved.

The physical collection of hardware repositories and functional units (Registers, ALU, Program Counter, and Memory) that data flows through as an instruction executes.

Every specific memory block in RAM maps to exactly one designated line in the cache. It is incredibly simple and cheap to build, but if a program frequently alternates between two different RAM blocks that map to the same cache line, it causes constant eviction conflicts (cache thrashing).

A distributed computing paradigm where data processing and storage are executed locally on decentralized physical devices (like smart sensors, cameras, or local gateways) at the "edge" of the network, rather than streaming raw data thousands of miles away to a centralized cloud datacenter.

A specific class of modern supercomputer capable of calculating at least 10^18 floating-point operations per second (1 exaFLOPS).