Amdahltétel

Amdahl's law, known in Hungarian as Amdahl-tétel, is a principle in parallel computing that describes the potential speedup of a task when portions of it can be executed concurrently. It was formulated by Gene M. Amdahl in 1967 to analyze the limits of performance gains from parallel processing.

The law states that if a fraction P of a program can be parallelized and the remaining fraction (1 − P) is inherently serial, then with N processors the overall speedup S is given by S = 1 / [(1 − P) + P / N]. The serial portion acts as a bottleneck, so as N increases the speedup approaches 1 / (1 − P). For example, if 90% of a task is parallelizable (P = 0.9), the maximum speedup with unlimited processors is 1 / 0.1 = 10.

Amdahl's law rests on several assumptions, including fixed problem size, identical processors, and negligible overhead from parallel execution. In real systems, communication, synchronization, and memory bandwidth constraints reduce the achievable speedup, making actual gains often well below the theoretical limit.

The law remains a foundational tool for performance modeling and architectural design, highlighting the importance of maximizing the parallelizable portion of workloads. It also motivates consideration of alternative perspectives, such as Gustafson's Law, which emphasizes scaling the problem size with more processors. Limitations of Amdahl's law include its emphasis on fixed workloads and its neglect of dynamic changes in workload or system heterogeneity.

See also: parallel computing, Amdahl's law, Gustafson's Law.