PFLOPS

PFLOPS stands for petaFLOPS, a unit of computing performance equal to one quadrillion floating-point operations per second (10^15). It is used to quantify the peak or sustained performance of high-performance computing systems on floating-point workloads.

FLOPS measure how quickly a computer can perform floating-point calculations. In practice, performance is reported as Rpeak (theoretical peak) and Rmax (sustained performance on a benchmark such as the High-Performance LINPACK, HPL). Real workloads often yield lower results due to memory bandwidth, communication, and software efficiency. PFLOPS provides a convenient scale when describing modern supercomputers, many of which use thousands or millions of processing units and accelerators.

Petascale computing emerged in the late 2000s. The first petaflop-class systems appeared around 2008–2010, with IBM Roadrunner commonly cited as the first to reach about 1 PFLOPS on LINPACK. Titan, a hybrid CPU–GPU system, reached roughly 17 PFLOPS in 2012. Tianhe-2 (Milky Way-2) surpassed 33 PFLOPS on LINPACK in 2013–2014, illustrating rapid scaling of performance.

Since then, the top supercomputers have progressed beyond PFLOPS into the petaflop and exaflop ranges, while PFLOPS remains a useful benchmark for describing performance and comparing systems at the scale of thousands of processors and accelerators. Modern discussions often distinguish between PFLOPS (petaFLOPS) as a performance scale and exaFLOPS (exaFLOPS) for the next level of growth, as energy efficiency and interconnects become critical factors.

Related topics include FLOPS, LINPACK benchmarks, high-performance computing, and exascale computing.