GIGAFLOPS

GigaFLOPS, short for giga floating-point operations per second, is a unit of computational speed used to express how many billions of floating-point operations a processor can perform each second. The term is commonly abbreviated as GFLOPS and is used for CPUs, GPUs, accelerators, and whole-system performance. One GFLOPS equals 1 x 10^9 floating-point operations per second.

FLOPS measure capability in peak theoretical terms; actual performance depends on the workload. In many definitions, a fused multiply-add (FMA) counts as two floating-point operations; however some benchmarks count it as a single operation. Peak GFLOPS can be estimated as the product of the number of floating-point cores, the clock frequency, and the number of operations per cycle. Real sustained GFLOPS are often substantially lower due to memory bandwidth, latency, and software efficiency.

GFLOPS gained prominence as processors progressed from millions to billions of floating-point operations per second. Early systems measured in megaflops to gigaflops; by the 2000s, GPUs and CPUs reported tens or hundreds of GFLOPS. In the 2010s and beyond, teraflops became common, but GFLOPS remains in use for historical comparisons and for smaller-scale systems.

Examples: a processor rated at 4 GFLOPS peak can perform up to four billion single-precision floating-point operations per second under ideal conditions. Depending on precision (single vs double) and instruction mix, the actual throughput varies. Benchmarks such as LINPACK or SPEC FP are used to estimate GFLOPS for comparability.

GFLOPS remains a widely used, though imperfect, metric for comparing computational speed across architectures, particularly when discussing raw arithmetic throughput rather than overall system performance.