MFLOP

MFLOP stands for million floating-point operations per second, a unit used to express the rate at which a computer can perform floating-point arithmetic. A floating-point operation is an arithmetic calculation performed on floating-point numbers, typically including addition, subtraction, multiplication, and division. Depending on counting conventions, a fused multiply-add (FMA) may be counted as one or two operations, which can affect reported MFLOP values.

Measurement of MFLOPS relies on benchmarks that specify which operations are counted and under what precision. Common benchmarks include linear algebra workloads and synthetic tests designed to stress floating-point units. Reported MFLOPS can reflect peak theoretical performance, real-world achieved performance, or specific workload performance, and results vary with data type (single vs double precision), compiler optimizations, and system architecture.

Historically, MFLOPS were widely used in the 1980s and 1990s to characterize the performance of vector and supercomputers. As computing power grew, the unit evolved into GFLOPS (billion FLOPS) and later higher orders, with TFLOPS (trillion FLOPS) becoming common for modern systems. MFLOPS remain cited in educational contexts and can appear in discussions of older hardware or specific embedded systems.

Practical considerations include the counting method and the impact of modern features such as fused operations, parallelism, and multi-core or many-core architectures, all of which influence how floating-point performance is reported. Users interpreting MFLOP figures should consider the benchmark, precision, and counting rules to make meaningful comparisons.