MFLOPS

MFLOPS, or million floating point operations per second, is a unit of computer performance used to express how quickly a system can perform floating-point arithmetic. It measures the number of floating-point operations a processor can complete in one second, scaled by one million.

Measurement and usage

MFLOPS is usually determined by benchmark programs that perform a large number of floating-point operations and record the elapsed time. The resulting value depends on the mix of operations (e.g., additions, multiplications, divisions), the precision (single vs double), and how the hardware and software exploit parallelism and vector units. It is a coarse metric, and real-world performance can differ due to memory bandwidth, cache behavior, and compiler optimizations.

Relation to other units

FLOPS stands for floating point operations per second, with MFLOPS representing millions of those per second. Larger scales include GFLOPS (billions) and TFLOPS (trillions). MFLOPS is most informative for workloads that are dominated by floating-point calculations but does not capture all aspects of system performance, such as memory latency or I/O.

Limitations

Because MFLOPS depends on the specific algorithm and implementation, it is not a universal measure of a system’s efficiency. Two processors with similar MFLOPS may perform very differently on real applications due to differences in architecture, memory hierarchy, and compiler support. Benchmark results should be interpreted in context and complemented with representative workloads.

See also: FLOPS, GFLOPS, TFLOPS, LINPACK.