What are gigaFLOPS (GFLOPS)?

A gigaFLOPS (GFLOPS) is a unit in computer science that’s equiv­a­lent to one billion floating point op­er­a­tions per second. FLOPS are used to measure the computing power of proces­sors, es­pe­cial­ly those used for sci­en­tif­ic ap­pli­ca­tions, machine learning and high-per­for­mance computers.

What are FLOPS and what are they used for?

FLOPS is short for floating point operations per second and indicates how many cal­cu­la­tions with floating point numbers a computer can perform in one second. Floating point op­er­a­tions are essential for different sci­en­tif­ic and technical ap­pli­ca­tions, as they make it possible to process both very large and very small numbers with high precision. They are essential in areas like physics, computer science and machine learning.

FLOPS are crucial for ap­pli­ca­tions that require complex cal­cu­la­tions, for example weather forecasts, climate modeling, the sim­u­la­tion of physical processes, machine learning and ar­ti­fi­cial in­tel­li­gence. Proces­sors with high FLOPS values can work quickly and ef­fi­cient­ly in those areas.

Special benchmark tests are performed to determine the FLOPS value of a processor. The tests evaluate how many floating point op­er­a­tions the system can complete in one second. Different types of cal­cu­la­tions are often included in order to get an accurate picture of the per­for­mance of a processor.

How many FLOPS are in one gigaFLOPS?

FLOPS units are based on the decimal system and thus cor­re­spond to powers of ten. One GFLOPS is 10⁹ FLOPS, meaning that a processor with the value of 1 GFLOPS can perform one billion (1,000,000,000) floating point op­er­a­tions per second.

What are the other FLOPS units?

In addition to GFLOPS, there are other FLOPS units that allow you to compare the per­for­mance of CPUs and GPUs. You can convert other FLOPS units into GFLOPS as follows:

How many FLOPS do modern computers and GPUs reach?

The computing power of modern computers, es­pe­cial­ly high-per­for­mance GPUs, has increased immensely in recent years. Modern graphics cards and super computers reach values ranging from several TFLOPS up to even exaFLOPS.

For example, NVIDIA H100 is a high-per­for­mance GPU for AI and data center ap­pli­ca­tions. It reaches peak per­for­mance at 989 teraFLOPS for FP32 Tensor Core cal­cu­la­tions. The NVIDIA A30, which was designed primarily for high per­for­mance computing and AI in­fer­ences, works at 10.3 teraFLOPS for FP32 cal­cu­la­tions.

The fastest su­per­com­put­er in the world, the Frontier, can now reach values of over one exaFLOPS, meaning it can process trillions of FLOPS without a problem.