In high-performance computing, what is a gigaflop, teraflop, or petaflop?

In high-performance computing, the terms gigaflop, teraflop, and petaflop are measurements of compute capacity describing the number of operations on floating-point numbers a system can perform in one second:

• Gigaflop: A gigaflop is one billion floating-point operations per second. "GigaFLOPS" is also sometimes used to indicate one billion floating-point operations per second.
  
  
• Teraflop: A teraflop is one trillion floating-point operations per second. "TeraFLOPS" is also sometimes used to indicate one trillion floating-point operations per second.
  
  
• Petaflop: A petaflop is one quadrillion floating-point operations per second (or one thousand teraflops). "PetaFLOPS" is also sometimes used to indicate one quadrillion floating-point operations per second.

  Note: Indiana University's Big Red II system has a peak performance of 1 petaflop. To match the number of calculations Big Red II can perform in one second, a human would have to perform one calculation every second for 31,687,535.79 years.

Computer vendors and resource providers typically list the computing capacity of their systems in terms of the number of gigaflops, teraflops, or petaflops achieved on standard benchmark programs, such as the LINPACK DP, TPP, and HPC benchmarks, and the SPEC integer and floating-point benchmarks.

The theoretical peak computing capacity of a system is obtained by multiplying the number of processors it comprises by the clock speed of the processors by the number of floating-point operations per second the processors are capable of performing.

This document was developed with support from National Science Foundation (NSF) grant OCI-1053575. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the NSF.