Floating point operations per second: Difference between revisions

Revision as of 19:29, 22 May 2004

Revision as of 21:32, 22 May 2004

Line 1:

In [[computing]], '''FLOPSflops''' (whichis standsan foracronym of ''[[floating point|'''fl'''oating point]]operations '''o'''perationsper second'''p'''er. '''s'''econd'')This is anused approximateas measurea ofmetric (with the flops as unit) for a [[computer]]'s processingperformance, speed.especially Becausein modernfields computersof can make many suchscientific calculations inthat amake givenheavy second,use the standardof [[SIfloating point]] prefixes are used: calculations.

* '''Megaflops''' means one million FLOPS

* '''Gigaflops''' means one billion FLOPS

* '''Teraflops''' means 10¹² FLOPS. The most powerful [[supercomputer]]s have speeds measured in '''teraflops'''.

* '''Petaflops''' means 10¹⁵ FLOPS, or a thousand teraflops.

One should speak in the singular of a ''flops'' and not of a ''flop'', although the latter is frequently encountered. The final ''s'' stands for ''second'' and does not indicate a plural.

It is interesting to note that the combined calculating power of all the computers on the planet is only several petaflops.

==The performance spectrum==

FLOPS, like [[million instructions per second|MIPS]], are arguably not very useful as a [[benchmark]] for modern computers because there are many other factors in computer performance other than raw floating-point computation speed, such as [[interprocessor communication]], [[cache coherence]], and the [[memory hierarchy]].

Computing devices exhibit an enormous range of performance levels in floating-point applications. Thus it makes sense to introduce larger units than the flops; the standard [[SI prefix|SI decimal prefix]]es are used for this purpose. For example, a cheap but modern desktop computer can make billions of floating point operations per second, so its performance is in the range of a few gigaflops (10⁹ flops).

Today's most powerful [[supercomputer]]s have speeds measured in teraflops (10¹² flops). The current record holder is [[Earth Simulator]], capable of 35 teraflops. The planned [[Blue Gene]] architecture may eventually reach speeds in excess of one petaflops (10¹⁵ slops). The most successful [[distributed computing]] projects are not far behind. For example, the [[GIMPS]] virtual computer runs at over 10 teraflops.

For ordinary (non-scientific) applications, [[integer]] operations (rather than floating point) are far more common. Measuring floating point operation speed, therefore, does not give an accurate measure of how the processor will perform. On the other hand, for scientific jobs, such as analysis of data, FLOPS measurement is effective.

[[Pocket calculator]]s are at the other end of the performance spectrum. Any response time below 0.1 second is experienced as 'instantaneous' by a human operator. Because it makes no sense to create a faster calculator, one may conclude that a pocket calculator performs at about 10 flops.

Of course, humans are even worse floating-point processors. If it takes a person a quarter of an hour to carry out a pencil-and-paper [[long division]] with 10 significant digits, that person would be calculating in the milliflops range.

==Flops as a metric==

In order for flops to be useful as a metric for floating-point performance, a standard benchmark must be available on all computers of interest. An example is the [[Linpack]] benchmark.

FLOPS,Flops likein [[million instructions per second|MIPS]],isolation are arguably not very useful as a [[benchmark]] for modern computers. because thereThere are many other factors in computer performance other than raw floating-point computation speed, such as [[interprocessor communication]], [[cache coherence]], and the [[memory hierarchy]].

For ordinary (non-scientific) applications, [[integer]] operations (rathermeasured thanin floating[[Million pointinstructions per second|MIPS]]) are far more common. Measuring floating point operation speed, therefore, does not givepredict an accurate measure ofaccurately how the processor will perform. Onon thejust otherany problem. handHowever, for many scientific jobs, such as analysis of data, FLOPSa flops measurementrating ''is'' effective.

[[ja:FLOPS]]