One petaFLOPS (PFLOPS) is equal to 1015 FLOPS or one quadrillion floating point op­er­a­tions per second. The unit is used to measure the power of high-per­for­mance computers.

What are FLOPS and what are they used for?

FLOPS stands for floating point operations per second and indicates how many floating point op­er­a­tions a computer can complete in one second. Floating point op­er­a­tions are essential for complex math­e­mat­i­cal and sci­en­tif­ic cal­cu­la­tions, as they make it possible to represent and process decimal numbers with high precision.

The per­for­mance of a computer in FLOPS is es­pe­cial­ly important for ap­pli­ca­tions that require immense computing capacity. That includes:

  • Sci­en­tif­ic sim­u­la­tions: In physics, chemistry and biology, FLOPS are used to calculate complex weather models, flow sim­u­la­tions and molecular dynamics.
  • Ar­ti­fi­cial in­tel­li­gence and machine learning: AI models and neural networks require immense computing power for training millions of pa­ra­me­ters and op­ti­miz­ing their per­for­mance.
  • Computer graphics and video games: High FLOPS values are par­tic­u­lar­ly crucial in 3D rendering and real-time cal­cu­la­tions for rep­re­sent­ing complex visual effects and physics sim­u­la­tions.

FLOPS are usually measured using stan­dard­ized bench­marks. One common test is the LINPACK benchmark, which evaluates how fast a computer solves linear systems of equations. That provides in­for­ma­tion about how efficient a system can work with floating point op­er­a­tions.

How many FLOPS is one PFLOPS?

One petaFLOPS is equal to 1015 or 1,000,000,000,000,000 floating point op­er­a­tions per second.

To get a feel for the immense computing power of a PFLOPS system: If each person on the planet (about 8 billion people) completed one cal­cu­la­tion per second, it would take 4 years to achieve what a petaFLOPS computer can do in a single second.

How to convert other FLOPS units into PFLOPS

In addition to PFLOPS, there are other FLOPS units that are used to describe the computing power of different systems. It’s easy to convert between these units, as shown in this table:

Unit FLOPS value Con­ver­sion into PFLOPS
KiloFLOP 103 FLOPS (1,000) 10-12 PFLOPS
MegaFLOP 106 FLOPS (1 million) 10-9 PFLOPS
GigaFLOP 109 FLOPS (1 billion) 10-6 PFLOPS
TeraFLOP 1012 FLOPS (1 trillion) 10-3 PFLOPS
PetaFLOP 1015 FLOPS (1 quadrillion) 1 PFLOPS
ExaFLOP 1018 FLOPS (1 quin­til­lion) 103 PFLOPS

How many FLOPS do modern computers reach?

With the in­creas­ing number of modern high-per­for­mance computers and the growing im­por­tance of high-per­for­mance computing, PFLOPS are becoming more and more important as a unit. One PFLOPS is equal to 1,000 teraFLOPS and marks the next level in the de­vel­op­ment of computing power.

While many modern GPUs are measured in teraFLOPS, a number of them are very close to PFLOPS. The NVIDIA H100, for example, delivers peak values of around 989 teraFLOPS for FP32 Tensor Core cal­cu­la­tions, which is almost 1 PFLOPS. On the other hand, the NVIDIA A30, which was designed for high-per­for­mance computing and AI reaches just 10.3 teraFLOPS, which is just a fraction of a PFLOPS.

Outside the realm of GPUs, the fastest su­per­com­put­ers in the world operate in the PFLOPS range and beyond. The Frontier su­per­com­put­er has reached values of over 1 exaFLOPS, which is more than 1,000 PFLOPS. Other systems like the Japanese su­per­com­put­er Fugaku or the earlier American su­per­com­put­er Summit also work in the PFLOPS range and are thus capable of solving very complex sci­en­tif­ic and technical problems.

Image: ION_CA_VPS_Fallback_960x320.png
Image: ION_CA_VPS_Fallback_1200x1200.png
Go to Main Menu