flop

FLOP stands for floating point operation, a basic arithmetic operation performed on numbers represented in floating point format. In computing, floating point operations include addition, subtraction, multiplication, and division, as well as more complex functions expressed in floating point arithmetic. The term is used as a unit of computational performance for hardware and software, especially in scientific computing and benchmarks.

Performance is expressed in FLOPS, floating point operations per second. Common prefixes denote scale: kiloFLOPS (10^3), megaFLOPS (10^6), gigaFLOPS (10^9), teraFLOPS (10^12), petaFLOPS (10^15), and exaFLOPS (10^18). A fused multiply-add (FMA) operation is typically counted as two FLOPs (one multiply and one add) in most benchmarks, though some methodologies treat an FMA as a single operation.

FLOPS-based measures have been central to high-performance computing since the 1960s. They are used to compare CPUs, GPUs, and supercomputers, informally ranking systems. Benchmarks such as LINPACK, High-Performance Linpack (HPL), and other suites report FLOPS to indicate computational capability; exascale systems target around 10^18 FLOPS.

Limitations: FLOPS do not capture memory bandwidth, latency, software efficiency, or algorithmic complexity. Real-world performance depends on many factors, including memory hierarchy, parallelization, and numerical stability. Consequently, FLOPS are best viewed as one metric among several when evaluating a system or algorithm.