Editing Floating point operations per second

{{short description|Measure of computer performance}}
{{Other uses|Flop (disambiguation){{!}}Flop}}
{{redirect-distinguish|Operations per second|Instructions per second}}
{{Use mdy dates|date=June 2022}}
'''Floating point operations per second''' ('''FLOPS''', '''flops''' or '''flop/s''') is a measure of [[computer performance]] in [[computing]], useful in fields of scientific computations that require [[floating-point]] calculations.<ref>{{cite web |title=Understand measures of supercomputer performance and storage system capacity |url=https://kb.iu.edu/d/apeq |website=kb.iu.edu |access-date=23 March 2024}}</ref> 

For such cases, it is a more accurate measure than measuring [[instructions per second]].{{cn|date=March 2024}}

==Floating-point arithmetic==
{{Anchor|multipliers}}
{| class="wikitable floatright sortable"
|+ Multipliers for flops
! Name
! Unit
! Value
|-
| [[Kilo-|kilo]]FLOPS
| kFLOPS
| 10<sup>3</sup>
|-
| [[Mega-|mega]]FLOPS
| MFLOPS
| 10<sup>6</sup>
|-
| [[Giga-|giga]]FLOPS
| GFLOPS
| 10<sup>9</sup>
|-
| [[Tera-|tera]]FLOPS
| TFLOPS
| 10<sup>12</sup>
|-
| [[Peta-|peta]]FLOPS
| PFLOPS
| 10<sup>15</sup>
|-
| [[Exa-|exa]]FLOPS
| EFLOPS
| 10<sup>18</sup>
|-
| [[Zetta-|zetta]]FLOPS
| ZFLOPS
| 10<sup>21</sup>
|-
| [[Yotta-|yotta]]FLOPS
| YFLOPS
| 10<sup>24</sup>
|-
| [[Ronna-|ronna]]FLOPS
| RFLOPS
|10<sup>27</sup>
|-
| [[Quetta-|quetta]]FLOPS
| QFLOPS
|10<sup>30</sup>
|-
|}
[[Floating-point arithmetic]] is needed for very large or very small [[real number]]s, or computations that require a large dynamic range. Floating-point representation is similar to scientific notation, except computers use [[Binary number|base two]] (with rare exceptions), rather than [[Decimal|base ten]]. The encoding scheme stores the sign, the [[exponent]] (in base two for Cray and [[VAX]], base two or ten for [[IEEE floating point]] formats, and base 16 for [[IBM hexadecimal floating-point|IBM Floating Point Architecture]]) and the [[significand]] (number after the [[radix point]]). While several similar formats are in use, the most common is [[IEEE 754-1985|ANSI/IEEE Std. 754-1985]]. This standard defines the format for 32-bit numbers called ''single precision'', as well as 64-bit numbers called ''double precision'' and longer numbers called ''extended precision'' (used for intermediate results). Floating-point representations can support a much wider range of values than fixed-point, with the ability to represent very small numbers and very large numbers.<ref>[http://www.dspguide.com/ch4/3.htm Floating Point] Retrieved on December 25, 2009.</ref>

===Dynamic range and precision===
The exponentiation inherent in floating-point computation assures a much larger dynamic range – the largest and smallest numbers that can be represented – which is especially important when processing data sets where some of the data may have extremely large range of numerical values or where the range may be unpredictable. As such, floating-point processors are ideally suited for computationally intensive applications.<ref>[http://www.analog.com/en/embedded-processing-dsp/content/Fixed-Point_vs_Floating-Point_DSP/fca.html Summary: Fixed-point (integer) vs floating-point] {{Webarchive|url=https://web.archive.org/web/20091231034504/http://www.analog.com/en/embedded-processing-dsp/content/Fixed-Point_vs_Floating-Point_DSP/fca.html |date=December 31, 2009 }} Retrieved on December 25, 2009.</ref>

===Computational performance===
FLOPS and [[Million instructions per second#Million instructions per second|MIPS]] are units of measure for the numerical computing performance of a computer. Floating-point operations are typically used in fields such as scientific computational research, as well as in [[machine learning]]. However, before the late 1980s floating-point hardware (it's possible to implement FP arithmetic in software over any integer hardware) was typically an optional feature, and computers that had it were said to be "scientific computers", or to have "[[scientific computation]]" capability. Thus the unit MIPS was useful to measure integer performance of any computer, including those without such a capability, and to account for architecture differences, similar MOPS (million operations per second) was used as early as 1970<ref>{{Cite book |url=https://books.google.com/books?id=ARwaAQAAMAAJ&pg=RA7-PA7 |title=NASA Technical Note |date=1970 |publisher=National Aeronautics and Space Administration. |language=en}}</ref> as well. Note that besides integer (or fixed-point) arithmetics, examples of integer operation include data movement (A to B) or value testing (If A = B, then C). That's why MIPS as a performance benchmark is adequate when a computer is used in database queries, word processing, spreadsheets, or to run multiple virtual operating systems.<ref>[http://www.dspguide.com/ch28/4.htm Fixed versus floating point.] Retrieved on December 25, 2009.</ref><ref>[http://www.dspguide.com/ch28/1.htm Data manipulation and math calculation.] Retrieved on December 25, 2009.</ref> In 1974 [[David Kuck]] coined the terms flops and megaflops for the description of supercomputer performance of the day by the number of floating-point calculations they performed per second.<ref>{{Cite book |last=Kuck |first=D. J. |url=https://books.google.com/books?id=PFqzjuXL4YgC&pg=PA1 |title=Computer System Capacity Fundamentals |date=1974 |publisher=U.S. Department of Commerce, National Bureau of Standards |language=en}}</ref> This was much better than using the prevalent MIPS to compare computers as this statistic usually had little bearing on the arithmetic capability of the machine on scientific tasks.

[[File:Supercomputer Power (FLOPS), OWID.svg|thumb|upright=1.35|FLOPS by the largest [[supercomputer]] over time]]
FLOPS on an HPC-system can be calculated using this equation:<ref name="en.community.dell.com">{{Cite web |url=https://www.dell.com/support/article/fr/fr/frbsdt1/sln310893/nodes-sockets-cores-and-flops-oh-my |title="Nodes, Sockets, Cores and FLOPS, Oh, My" by Dr. Mark R. Fernandez, Ph.D. |access-date=February 12, 2019 |archive-date=February 13, 2019 |archive-url=https://web.archive.org/web/20190213005604/https://www.dell.com/support/article/fr/fr/frbsdt1/sln310893/nodes-sockets-cores-and-flops-oh-my |url-status=dead }}</ref>

: <math>\text{FLOPS} = \text{racks} \times \frac{\text{nodes}}{\text{rack}} \times \frac{\text{sockets}}{\text{node}} \times \frac{\text{cores}}{\text{socket}} \times \frac{\text{cycles}}{ \text{second}} \times \frac{\text{FLOPs}}{\text{cycle}}.</math>

This can be simplified to the most common case: a computer that has exactly 1 CPU:

: <math>\text{FLOPS} = \text{cores} \times \frac{\text{cycles}}{ \text{second}} \times \frac{\text{FLOPs}}{\text{cycle}}.</math>

FLOPS can be recorded in different measures of precision, for example, the [[TOP500]] supercomputer list ranks computers by 64-bit ([[double-precision floating-point format]]) operations per second, abbreviated to ''FP64''.<ref name="top500faq">{{cite web |title=FREQUENTLY ASKED QUESTIONS |url=https://www.top500.org/resources/frequently-asked-questions/ |website=top500.org |access-date=June 23, 2020}}</ref> Similar measures are available for [[Single-precision floating-point format|32-bit]] (''FP32'') and [[Half-precision floating-point format|16-bit]] (''FP16'') operations.

{{anchor|FLOPSforProcessors}}

== Floating-point operations per clock cycle for various processors ==
{{Table alignment}}
{{sort-under}}
{| class="wikitable sortable sort-under col3right col4right col5right"
|+ Floating-point operations per clock cycle per core<ref>{{Cite web | url=https://en.wikichip.org/wiki/flops | title=Floating-Point Operations Per Second (FLOPS)}}</ref>
! scope="col" | Microarchitecture
! scope="col" | [[Instruction set architecture]]
! scope="col" | FP64
! scope="col" | FP32
! scope="col" | FP16
|-
! colspan="5" |Intel CPU
|-
|[[Intel 80486]]
|[[x87]] (32-bit)
| {{dunno}}
|0.128<ref name=":1" />
| {{dunno}}
|-
|{{plainlist|
*Intel [[P5 (microarchitecture)|P5]] [[Pentium]]
*Intel [[P6 (microarchitecture)|P6]] [[Pentium Pro]]
}}
|[[x87]] (32-bit)
| {{dunno}}
|0.5<ref name=":1">{{Cite web|title=home.iae.nl |url=http://home.iae.nl/users/mhx/flops_4.tbl|access-date=|website=}}</ref>
| {{dunno}}
|-
|{{plainlist|
*Intel [[P5 (microarchitecture)|P5]] [[Pentium]] MMX
*Intel [[P6 (microarchitecture)|P6]] [[Pentium II]]
}}
|[[MMX (instruction set)|MMX]] (64-bit)
| {{dunno}}
|1<ref name=":0">{{Cite web|title=Computing Power throughout History|url=https://www.alternatewars.com/BBOW/Computing/Computing_Power.htm|access-date=2021-02-13|website=alternatewars.com}}</ref>
| {{dunno}}
|-
|Intel [[P6 (microarchitecture)|P6]] [[Pentium III]]
|[[Streaming SIMD Extensions|SSE]] (64-bit)
| {{dunno}}
|2<ref name=":0" />
| {{dunno}}
|-
|Intel [[NetBurst]] [[Pentium 4]] (Willamette, Northwood)
|[[SSE2]] (64-bit)
|2
|4
| {{dunno}}
|-
|Intel [[P6 (microarchitecture)|P6]] [[Pentium M]]
|[[SSE2]] (64-bit)
|1
|2
| {{dunno}}
|-
|{{plainlist|
*Intel [[NetBurst]] [[Pentium 4]] (Prescott, Cedar Mill)
*Intel [[NetBurst]] [[Pentium D]] (Smithfield, Presler)
*Intel [[P6 (microarchitecture)|P6]] [[Intel Core|Core]] ([[Yonah (microprocessor)|Yonah]])
}}
|[[SSE3]] (64-bit)
|2
|4
| {{dunno}}
|-
|{{plainlist|
*Intel [[Intel Core (microarchitecture)|Core]] ([[Merom (microprocessor)|Merom]], [[Penryn (microarchitecture)|Penryn]])
*Intel [[Nehalem (microarchitecture)|Nehalem]]<ref name="tpeak_jos">{{Cite journal| title=Theoretical Peak FLOPS per instruction set: a tutorial | first1 = Romain | last1 = Dolbeau | year = 2017 |journal=Journal of Supercomputing |volume=74 |issue=3 |pages=1341–1377 |doi=10.1007/s11227-017-2177-5 | s2cid = 3540951 }}</ref> ([[Nehalem (microarchitecture)|Nehalem]], [[Westmere (microarchitecture)|Westmere]])
}}
|{{plainlist|
*[[SSSE3]] (128-bit)
*[[SSE4]] (128-bit)
}}
| 4 || 8 || {{dunno}}
|-
| Intel [[Atom (system on chip)|Atom]] ([[Bonnell (microarchitecture)|Bonnell]], [[Saltwell (microarchitecture)|Saltwell]], [[Silvermont (microarchitecture)|Silvermont]] and [[Goldmont]]) || [[SSE3]] (128-bit) 
| 2 || 4 || {{dunno}}
|-
| Intel [[Sandy Bridge]] ([[Sandy Bridge]], [[Ivy Bridge (microarchitecture)|Ivy Bridge]]) || [[Advanced Vector Extensions|AVX]] (256-bit) 
| 8 || 16 || 0
|-
|{{ublist|
| Intel [[Haswell (microarchitecture)|Haswell]]<ref name="tpeak_jos"/> ([[Haswell (microarchitecture)|Haswell]], [[Haswell (microarchitecture)|Devil's Canyon]], [[Broadwell (microarchitecture)|Broadwell]])
| Intel [[Skylake (microarchitecture)|Skylake]]<br/>([[Skylake (microarchitecture)|Skylake]], [[Kaby Lake]], [[Coffee Lake]], [[Comet Lake (microprocessor)|Comet Lake]], [[Whiskey Lake (microarchitecture)|Whiskey Lake]], [[Amber Lake (microarchitecture)|Amber Lake]])
}}
|[[Advanced Vector Extensions|AVX2]] & [[FMA instruction set|FMA]] (256-bit) 
| 16 || 32 || 0
|-
| Intel [[Xeon Phi]] ([[Knights Corner]]) || [[Initial Many Core Instructions|IMCI]] (512-bit) 
| 16 || 32 || 0
|-
|{{plainlist|
* Intel [[Skylake (microarchitecture)|Skylake-X]] ([[Skylake (microarchitecture)|Skylake-X]], [[Cascade Lake (microarchitecture)|Cascade Lake]])
* Intel [[Xeon Phi]] ([[Knights Landing (microarchitecture)|Knights Landing]], [[Knights Mill]])
* Intel [[Ice Lake (microprocessor)|Ice Lake]], [[Tiger Lake (microprocessor)|Tiger Lake]] and [[Rocket Lake]]
}}
| [[Advanced Vector Extensions|AVX-512]] & [[FMA instruction set|FMA]] (512-bit) 
| 32 || 64 || 0
|-
! colspan="5" |AMD CPU
|-
| AMD [[Bobcat (microarchitecture)|Bobcat]] || [[x86-64|AMD64]] (64-bit) || 2 || 4 || 0
|-
|{{plainlist|
*AMD [[Jaguar (microarchitecture)|Jaguar]]
*AMD [[Puma (microarchitecture)|Puma]]
}}
|[[Advanced Vector Extensions|AVX]] (128-bit) 
| 4 || 8 || 0
|-
|AMD [[AMD 10h|K10]]
|[[SSE4|SSE4/4a]] (128-bit) || 4 || 8 || 0
|-
| AMD [[Bulldozer (microarchitecture)|Bulldozer]]<ref name="tpeak_jos" /><br/>([[Piledriver (microarchitecture)|Piledriver]], [[Steamroller (microarchitecture)|Steamroller]], [[Excavator (microarchitecture)|Excavator]])
|{{ublist|
|[[Advanced Vector Extensions|AVX]] (128-bit)<br/>(Bulldozer, Steamroller)
|[[AVX2]] (128-bit) (Excavator)
|[[FMA instruction set|FMA3]] (Bulldozer)<ref>{{Cite web|url=https://developer.amd.com/wordpress/media/2012/10/New-Bulldozer-and-Piledriver-Instructions.pdf|title=New instructions support for Bulldozer (FMA3) and Piledriver (FMA3+4 and CVT, BMI, TB M)}}</ref>
|[[FMA instruction set|FMA3/4]] (Piledriver, Excavator)
}}
| 4 || 8 || 0
|-
|{{ublist|
|AMD [[Zen (microarchitecture)|Zen]]<br/>(Ryzen 1000 series, Threadripper 1000 series, Epyc [[Epyc|Naples]])
|AMD [[Zen+]]<ref name="tpeak_jos"/><ref>{{Cite web | url=http://www.agner.org/optimize/blog/read.php?i=838 | title=Agner's CPU blog - Test results for AMD Ryzen}}</ref><ref>https://arstechnica.com/gadgets/2017/03/amds-moment-of-zen-finally-an-architecture-that-can-compete/2/ "each core now has a pair of 128-bit FMA units of its own"</ref><ref>{{cite conference |url=https://www.hotchips.org/wp-content/uploads/hc_archives/hc28/HC28.23-Tuesday-Epub/HC28.23.90-High-Perform-Epub/HC28.23.930-X86-core-MikeClark-AMD-final_v2-28.pdf#page=7 |title=A New x86 Core Architecture for the Next Generation of Computing |author=Mike Clark |date=August 23, 2016 |publisher=AMD |conference=HotChips 28 |access-date=October 8, 2017 |archive-date=July 31, 2020 |archive-url=https://web.archive.org/web/20200731171730/https://www.hotchips.org/wp-content/uploads/hc_archives/hc28/HC28.23-Tuesday-Epub/HC28.23.90-High-Perform-Epub/HC28.23.930-X86-core-MikeClark-AMD-final_v2-28.pdf#page=7 |url-status=dead }} [https://images.anandtech.com/doci/10591/HC28.AMD.Mike%20Clark.final-page-007.jpg page 7]</ref><br/>(Ryzen 2000 series, Threadripper 2000 series)
}}
| [[Advanced Vector Extensions|AVX2]] & [[FMA instruction set|FMA]]<br/>(128-bit, 256-bit decoding)<ref>{{Cite web |title=The microarchitecture of Intel and AMD CPUs |url=https://www.agner.org/optimize/microarchitecture.pdf}}</ref> 
| 8 || 16 || 0
|-
|{{ublist|
|AMD [[Zen 2]]<ref name="www.youtube.com">{{cite web |url=https://www.youtube.com/watch?v=_96stDCb-mk&t=3299 |title=AMD CEO Lisa Su's COMPUTEX 2019 Keynote |archive-url=https://ghostarchive.org/varchive/youtube/20211211/_96stDCb-mk| archive-date=2021-12-11 |url-status=live |website=youtube.com|date=May 27, 2019 }}{{cbignore}}</ref><br/>(Ryzen 3000 series, Threadripper 3000 series, Epyc [[Epyc|Rome]])
|AMD [[Zen 3]]<br/>(Ryzen 5000 series, Epyc [[Epyc|Milan]])
}}
| [[Advanced Vector Extensions|AVX2]] & [[FMA instruction set|FMA]] (256-bit) 
| 16 || 32 || 0
|-
! colspan="5" |ARM CPU
|-
| ARM Cortex-A7, A9, A15 || [[ARM architecture|ARMv7]] 
| 1 || 8 || 0
|-
| ARM Cortex-A32, A35 || [[ARM architecture|ARMv8]] 
| 2 || 8 || 0
|-
| [[ARM Cortex-A53]], [[ARM Cortex-A55|A55]], [[ARM Cortex-A57|A57]],<ref name="tpeak_jos"/> [[ARM Cortex-A72|A72]], [[ARM Cortex-A73|A73]], [[ARM Cortex-A75|A75]] || [[ARM architecture|ARMv8]] 
| 4 || 8 || 0
|-
| [[ARM Cortex-A76]], [[ARM Cortex-A77|A77]], [[ARM Cortex-A78|A78]]|| [[ARM architecture|ARMv8]] 
| 8 || 16 || 0
|-
| [[ARM Cortex-X1]] || [[ARM architecture|ARMv8]] 
| 16 || 32 || {{dunno}}
|-
| Qualcomm [[Krait (CPU)|Krait]] || [[ARM architecture|ARMv8]] 
| 1 || 8 || 0
|-
| Qualcomm [[Kryo]] (1xx - 3xx) || [[ARM architecture|ARMv8]] 
| 2 || 8 || 0
|-
| Qualcomm [[Kryo]] (4xx - 5xx) || [[ARM architecture|ARMv8]] 
| 8 || 16 || 0
|-
| Samsung [[Exynos]] M1 and M2 || [[ARM architecture|ARMv8]] 
| 2 || 8 || 0
|-
| Samsung [[Exynos]] M3 and M4 || [[ARM architecture|ARMv8]] 
| 3 || 12 || 0
|-
| IBM PowerPC [[IBM A2|A2]] (Blue Gene/Q) || {{dunno}} 
| 8 || 8<br/>(as FP64) || 0
|-
| [[Hitachi SH-4]]<ref>{{cite journal |title=Entertainment Systems and High-Performance Processor SH-4 |journal=Hitachi Review |date=1999 |volume=48 |issue=2 |pages=58–63 |publisher=[[Hitachi]] |url=https://retrocdn.net/images/f/fa/Entertainment_Systems_and_High-Performance_Processor_SH-4.pdf |access-date=June 21, 2019}}</ref><ref>{{cite web |title=SH-4 Next-Generation DSP Architecture for VoIP |url=https://retrocdn.net/images/b/b3/SH-4_Next-Generation_DSP_Architecture.pdf |publisher=[[Hitachi]] |year=2000 |access-date=June 21, 2019}}</ref> || [[Hitachi SH-4|SH-4]] 
| 1 || 7 || 0
|-
! colspan="5" |Nvidia GPU
|-
|Nvidia [[Curie (microarchitecture)|Curie]] ([[GeForce 6 series]] and [[GeForce 7 series]])
|[[Parallel Thread Execution|PTX]] || {{dunno}} || 8 || {{dunno}}
|-
|Nvidia [[Tesla (microarchitecture)|Tesla]] 2.0 (GeForce GTX 260–295)
|[[Parallel Thread Execution|PTX]] || {{dunno}} || 2 || {{dunno}}
|-
| Nvidia [[Fermi (microarchitecture)|Fermi]]
(only GeForce GTX 465–480, 560 Ti, 570–590)
| [[Parallel Thread Execution|PTX]] 
| {{1/4}}<br/>(locked by driver,<br/>1 in hardware) || 2 || 0
|-
| Nvidia [[Fermi (microarchitecture)|Fermi]]
(only Quadro 600–2000)
| [[Parallel Thread Execution|PTX]] 
| {{frac|1|8}} || 2 || 0
|-
| Nvidia [[Fermi (microarchitecture)|Fermi]]
(only Quadro 4000–7000, Tesla)
| [[Parallel Thread Execution|PTX]] 
| 1 || 2 || 0
|-
| Nvidia [[Kepler (microarchitecture)|Kepler]]
(GeForce (except Titan and Titan Black), Quadro (except K6000), Tesla K10)
| [[Parallel Thread Execution|PTX]] 
| {{frac|1|12}}<br/>(for [[GeForce 700 series|GK110]]:<br/>locked by driver,<br/>{{2/3}} in hardware) || 2 || 0
|-
| Nvidia [[Kepler (microarchitecture)|Kepler]]
(GeForce GTX Titan and Titan Black, Quadro K6000, Tesla (except K10))
| [[Parallel Thread Execution|PTX]] 
| {{2/3}} || 2 || 0
|-
|{{ublist|
| Nvidia [[Maxwell (microarchitecture)|Maxwell]]
| Nvidia [[Pascal (microarchitecture)|Pascal]]<br/>(all except Quadro GP100 and Tesla P100)
}}
| [[Parallel Thread Execution|PTX]] || {{frac|1|16}} || 2 || {{frac|1|32}}
|-
| Nvidia [[Pascal (microarchitecture)|Pascal]] (only Quadro GP100 and Tesla P100) || [[Parallel Thread Execution|PTX]] || 1 || 2 || 4
|-
| Nvidia [[Volta (microarchitecture)|Volta]]<ref name="Nvidia Volta">{{cite web|title=Inside Volta: The World's Most Advanced Data Center GPU|date=May 10, 2017 |url=https://devblogs.nvidia.com/inside-volta/}}</ref> || [[Parallel Thread Execution|PTX]] 
| 1 || 2&nbsp;([[FP32]]) + 2&nbsp;([[Int32|INT32]]) || 16
|-
| Nvidia [[Turing (microarchitecture)|Turing]] (only GeForce [[GeForce 16 series|16XX]]) || [[Parallel Thread Execution|PTX]] 
| {{frac|1|16}} || 2&nbsp;(FP32) + 2&nbsp;(INT32) || 4
|-
| Nvidia [[Turing (microarchitecture)|Turing]] (all except GeForce [[GeForce 16 series|16XX]]) || [[Parallel Thread Execution|PTX]] 
| {{frac|1|16}} || 2&nbsp;(FP32) + 2&nbsp;(INT32) || 16
|-
| Nvidia [[Ampere (microarchitecture)|Ampere]]<ref name="Nvidia Ampere 1">{{cite web|title=NVIDIA Ampere Architecture In-Depth |date=May 14, 2020 |url=https://devblogs.nvidia.com/nvidia-ampere-architecture-in-depth/}}</ref><ref name="Nvidia Ampere 2">{{Cite web |title=NVIDIA A100 GPUs Power the Modern Data Center |website=NVIDIA |url=https://www.nvidia.com/en-us/data-center/a100/}}</ref> (only Tesla A100/A30) || [[Parallel Thread Execution|PTX]] 
| 2 || 2&nbsp;(FP32) + 2&nbsp;(INT32) || 32
|-
|{{plainlist|
* Nvidia [[Ampere (microarchitecture)|Ampere]] (all GeForce and Quadro, Tesla A40/A10)
* Nvidia [[Ada Lovelace (microarchitecture)|Ada Lovelace]]
}}
| [[Parallel Thread Execution|PTX]] || {{frac|1|32}} || {{nowrap|2&nbsp;(FP32) + 0&nbsp;(INT32)}}<br/>''or''<br/>{{nowrap|1&nbsp;(FP32) + 1&nbsp;(INT32)}} || 8
|-
| Nvidia [[Hopper (microarchitecture)|Hopper]] || [[Parallel Thread Execution|PTX]] || 2 || 2&nbsp;(FP32) + 1&nbsp;(INT32) || 32
|-
! colspan="5" |AMD GPU
|-
| AMD [[TeraScale (microarchitecture)#TeraScale 1|TeraScale 1]] ([[Radeon HD 4000 series]])
|[[TeraScale (microarchitecture)#TeraScale 1|TeraScale 1]] || 0.4 || 2 || {{dunno}}
|-
| AMD [[TeraScale (microarchitecture)#TeraScale 2|TeraScale 2]] ([[Radeon HD 5000 series]])
|[[TeraScale (microarchitecture)#TeraScale 2|TeraScale 2]] || 1 || 2 || {{dunno}}
|-
| AMD [[TeraScale (microarchitecture)#TeraScale 3|TeraScale 3]] ([[Radeon HD 6000 series]])
|[[TeraScale (microarchitecture)#TeraScale 3|TeraScale 3]] || 1 || 4 || {{dunno}}
|-
| AMD [[Graphics Core Next|GCN]]<br/>(only Radeon Pro W 8100–9100)
| [[Graphics Core Next|GCN]] || 1 || 2 || {{dunno}}
|-
| AMD [[Graphics Core Next|GCN]]<br/>(all except Radeon Pro W 8100–9100, Vega 10–20)
| [[Graphics Core Next|GCN]] || {{frac|1|8}} || 2 || 4
|-
| AMD [[AMD RX Vega series|GCN Vega 10]] || [[Graphics Core Next|GCN]] || {{frac|1|8}} || 2 || 4
|-
| AMD [[AMD RX Vega series|GCN Vega 20]]<br/>(only Radeon VII) || [[Graphics Core Next|GCN]] 
| {{1/2}}<br/>(locked by driver,<br/>1 in hardware) || 2 || 4
|-
| AMD [[AMD RX Vega series|GCN Vega 20]]<br/>(only Radeon Instinct MI50 / MI60 and Radeon Pro VII)
| [[Graphics Core Next|GCN]] 
| 1 || 2 || 4
|-
|{{plainlist|
* AMD [[AMD Radeon RX 5000 series|RDNA]]<ref name="hardwareluxx">{{Cite web|url=https://www.hardwareluxx.de/index.php/artikel/hardware/grafikkarten/49892-alles-zu-navi-radeon-rx-5700-xt-ist-rdna-mit-gddr6.html|title=Die RDNA-Architektur - Seite 2|first=Andreas|last=Schilling|website=Hardwareluxx|date=June 10, 2019 }}</ref><ref name="techpowerup">{{Cite web|url=https://www.techpowerup.com/gpu-specs/radeon-rx-5700-xt.c3339|title=AMD Radeon RX 5700 XT Specs|website=TechPowerUp}}</ref>
* AMD [[RDNA 2]]
}}
| [[AMD RDNA Architecture|RDNA]] 
| {{frac|1|8}} || 2 || 4
|-
| AMD RDNA3 || [[AMD RDNA Architecture|RDNA]] 
| {{frac|1|8}}? || 4 || 8?
|-
| AMD [[AMD CDNA Architecture|CDNA]] || [[AMD CDNA Architecture|CDNA]] 
| 1 || 4<br/>(Tensor)<ref name="AMD">{{cite web|url=https://www.amd.com/en/products/server-accelerators/instinct-mi100|title=AMD Instinct MI100 Accelerator}}</ref> || 16
|-
| AMD [[AMD CDNA Architecture|CDNA 2]] || [[AMD CDNA Architecture|CDNA 2]] 
| 4<br/>(Tensor) || 4<br/>(Tensor) || 16
|-
! colspan="5" |Intel GPU
|-
| Intel Xe-LP (Iris Xe MAX)<ref name="intel.com">{{cite web|url=https://www.intel.com/content/www/us/en/developer/articles/technical/introduction-to-the-xe-hpg-architecture.html|title=Introduction to the Xe-HPG Architecture}}</ref> || Xe 
| {{1/2}}? || 2 || 4
|-
| Intel Xe-HPG (Arc Alchemist)<ref name="intel.com"/> || Xe || 0 || 2 || 16
|-
| Intel Xe-HPC (Ponte Vecchio)<ref name="allinfo.space">{{cite web|url=https://allinfo.space/2022/11/09/intel-data-center-gpu-max-ponte-vecchio-starts-in-3-variants-for-supercomputers/|title=Intel Data Center GPU Max|date=November 9, 2022 }}</ref> || Xe || 2 || 2 || 32
|-
| Intel Xe2 (Arc Battlemage) || Xe2 || {{frac|1|8}} || 2 || 16
|-
! colspan="5" |Qualcomm GPU
|-
|Qualcomm [[Adreno]] 5x0
|[[Adreno]] 5xx || 1 || 2 || 4
|-
|Qualcomm [[Adreno]] 6x0
|[[Adreno]] 6xx || 1 || 2 || 4
|-
! colspan="5" |Graphcore
|-
| Graphcore Colossus GC2<ref name="Source 2">{{cite web|url=https://www.youtube.com/watch?v=2IOyQEIlN6Y&t=1361|title=250 TFLOPs/s for two chips with FP16 mixed precision|website=youtube.com|date=October 26, 2018 }}</ref><ref name="Source 3">Archived at [https://ghostarchive.org/varchive/youtube/20211211/7XtBZ4Hsi_M Ghostarchive]{{cbignore}} and the [https://web.archive.org/web/20180119094342/https://www.youtube.com/watch?v=7XtBZ4Hsi_M&gl=US&hl=en Wayback Machine]{{cbignore}}: {{cite web|url=https://www.youtube.com/watch?v=7XtBZ4Hsi_M&t=2208|title=Estimation via power consumption that FP32 is 1/4 of FP16 and that clock frequency is below 1.5GHz|website=youtube.com|date=October 25, 2017 }}{{cbignore}}</ref> 
| {{dunno}} || 0 || 16 || 64
|-
|{{plainlist|
* Graphcore Colossus GC200 Mk2<ref name="Source 4">Archived at [https://ghostarchive.org/varchive/youtube/20211211/_zvU0uwIafQ Ghostarchive]{{cbignore}} and the [https://web.archive.org/web/20200716143430/https://www.youtube.com/watch?v=_zvU0uwIafQ Wayback Machine]{{cbignore}}: {{cite web|url=https://www.youtube.com/watch?v=_zvU0uwIafQ|title=Introducing Graphcore's Mk2 IPU systems|website=youtube.com|date=July 15, 2020 }}{{cbignore}}</ref>

*Graphcore Bow-2000<ref name="Source 5">{{cite web|url=https://docs.graphcore.ai/projects/bow-2000-datasheet/en/latest/product-description.html#technical-specifications|title=Bow-2000 IPU-Machine|website=docs.graphcore.ai/}}{{cbignore}}</ref>
}}
| {{dunno}} || 0 || 32 || 128
|-
! colspan="5" |[[Supercomputer]]
|-
|[[ENIAC]] @ 100&nbsp;kHz in 1945
|
|0.004<ref>ENIAC @ 100&nbsp;kHz with 385 Flops {{Cite web|title=Computers of Yore|url=https://www.clear.rice.edu/comp201/08-spring/lectures/lec02/computers.shtml|access-date=2021-02-26|website=clear.rice.edu}}</ref><br/>(~{{val|3|e=-8|u=FLOPS|upl=W}})
|
|
|-
|48-bit processor @ 208 [[Kilohertz|kHz]] in [[CDC 1604]] in 1960
|
|
|
|
|-
|60-bit processor @ 10 MHz in [[CDC 6600]] in 1964
|
|0.3<br/>(FP60) 
|
|
|-
|60-bit processor @ 10 MHz in [[CDC 7600]] in 1967
|
|1.0<br/>(FP60)
|
|
|-
|[[Cray-1]] @ 80 MHz in 1976
|
|2<br/>(700 FLOPS/W)
|
|
|-
|[[CDC Cyber]] 205 @ 50 MHz in 1981
[[Fortran|FORTRAN]] compiler 
(ANSI 77 with vector extensions)
|
|8
|16
|
|-
|[[Transputer]] IMS T800-20 @ 20 MHz in 1987
|
|0.08<ref name="INMOSTN6 (1988)">{{Cite web|title=IMS T800 Architecture|url=https://www.transputer.net/tn/06/tn06.html#x1-150005|access-date=2023-12-28|website=transputer.net}}</ref>
|
|
|-
|[[Parallella]] E16 @ 1000 MHz in 2012
|
|2<ref name="Epiphany multi-core coprocessor E16G301 specs">[http://www.adapteva.com/products/silicon-devices/e16g301/ Epiphany-III 16-core 65nm Microprocessor (E16G301)] // [http://www.adapteva.com/author/admin/ admin] (August 19, 2012)</ref><br/>(5.0&nbsp;GFLOPS/W)<ref name="FeldmanM_(2014)"/>
|
|
|-
|[[Parallella]] E64 @ 800 MHz in 2012
|
|2<ref name="Epiphany multi-core coprocessor E64G401 specs">[http://www.adapteva.com/products/silicon-devices/e64g401/ Epiphany-IV 64-core 28nm Microprocessor (E64G401)] // [http://www.adapteva.com/author/admin/ admin] (August 19, 2012)</ref><br/>(50.0&nbsp;GFLOPS/W)<ref name="FeldmanM_(2014)">{{cite web|url=http://www.hpcwire.com/2012/08/22/adapteva_unveils_64-core_chip/|title=Adapteva Unveils 64-Core Chip|last= Feldman|first=Michael|date=August 22, 2012|publisher=HPCWire|accessdate=September 3, 2014}}</ref>
|
|
|-
!Microarchitecture
![[Instruction set architecture]]
!FP64
!FP32
!FP16
|}

==Performance records==
===Single computer records===
In June 1997, [[Intel]]'s [[ASCI Red]] was the world's first computer to achieve one teraFLOPS and beyond. Sandia director Bill Camp said that ASCI Red had the best reliability of any supercomputer ever built, and "was supercomputing's high-water mark in longevity, price, and performance".<ref name="jacobsequity.com">{{cite web |title=Sandia's ASCI Red, world's first teraflop supercomputer, is decommissioned |url=http://www.jacobsequity.com/ASCI%20Red%20Supercomputer.pdf |access-date=November 17, 2011 |archive-url=https://web.archive.org/web/20101105131112/http://www.jacobsequity.com/ASCI%20Red%20Supercomputer.pdf |archive-date=November 5, 2010 }}</ref>

[[NEC]]'s [[NEC SX-9|SX-9]] supercomputer was the world's first [[vector processor]] to exceed 100&nbsp;gigaFLOPS per single core.

In June 2006, a new computer was announced by Japanese research institute [[RIKEN]], the [[MDGRAPE-3]]. The computer's performance tops out at one petaFLOPS, almost two times faster than the Blue Gene/L, but MDGRAPE-3 is not a general purpose computer, which is why it does not appear in the [[TOP500|Top500.org]] list. It has special-purpose [[pipeline (computing)|pipelines]] for simulating molecular dynamics.

By 2007, [[Intel|Intel Corporation]] unveiled the experimental [[multi-core]] [[Teraflops Research Chip|POLARIS]] chip, which achieves 1 teraFLOPS at 3.13&nbsp;GHz. The 80-core chip can raise this result to 2 teraFLOPS at 6.26&nbsp;GHz, although the thermal dissipation at this frequency exceeds 190&nbsp;watts.<ref>{{cite web|author=Richard Swinburne |url=http://www.bit-tech.net/hardware/2007/04/30/the_arrival_of_teraflop_computing/2 |title=The Arrival of TeraFLOP Computing |publisher=bit-tech.net |date=April 30, 2007 |access-date=February 9, 2012}}</ref>

In June 2007, Top500.org reported the fastest computer in the world to be the [[Blue Gene|IBM Blue Gene/L]] supercomputer, measuring a peak of 596&nbsp;teraFLOPS.<ref>{{cite news |url=http://top500.org/news/2007/06/23/29th_top500_list_world_s_fastest_supercomputers_released |title=29th TOP500 List of World's Fastest Supercomputers Released |date=June 23, 2007 |website=Top500.org |access-date=July 8, 2008 |archive-url=https://web.archive.org/web/20080509064814/http://www.top500.org/news/2007/06/23/29th_top500_list_world_s_fastest_supercomputers_released |archive-date=May 9, 2008 |df=mdy }}</ref> The [[Cray XT4]] hit second place with 101.7&nbsp;teraFLOPS.

On June 26, 2007, [[IBM]] announced the second generation of its top supercomputer, dubbed Blue Gene/P and designed to continuously operate at speeds exceeding one petaFLOPS, faster than the Blue Gene/L. When configured to do so, it can reach speeds in excess of three petaFLOPS.<ref>{{cite news |url=http://www.top500.org/lists/2008/06 |title=June 2008 |publisher=TOP500 |access-date=July 8, 2008 }}</ref>

On October 25, 2007, [[NEC]] Corporation of Japan issued a press release announcing its SX series model [[SX-9]],<ref>{{cite news|url=http://www.nec.co.jp/press/en/0710/2501.html|title=NEC Launches World's Fastest Vector Supercomputer, SX-9|date=October 25, 2007|publisher=NEC|access-date=July 8, 2008}}</ref> claiming it to be the world's fastest vector supercomputer. The [[SX-9]] features the first CPU capable of a peak vector performance of 102.4 gigaFLOPS per single core.

On February 4, 2008, the [[National Science Foundation|NSF]] and the [[University of Texas at Austin]] opened full scale research runs on an [[AMD]], [[Sun Microsystems|Sun]] supercomputer named Ranger,<ref>{{cite web
 |url          = http://www.tacc.utexas.edu/resources/hpcsystems/
 |title        = University of Texas at Austin, Texas Advanced Computing Center
 |access-date  = September 13, 2010
 |quote        = Any researcher at a U.S. institution can submit a proposal to request an allocation of cycles on the system.
 |archive-url  = https://web.archive.org/web/20090801102108/http://www.tacc.utexas.edu/resources/hpcsystems/
 |archive-date = August 1, 2009
 |df           = mdy
}}</ref>
the most powerful supercomputing system in the world for open science research, which operates at sustained speed of 0.5&nbsp;petaFLOPS.

On May 25, 2008, an American supercomputer built by [[IBM]], named '[[IBM Roadrunner|Roadrunner]]', reached the computing milestone of one petaFLOPS. It headed the June 2008 and November 2008 [[TOP500]] list of the most powerful supercomputers (excluding [[grid computing|grid computers]]).<ref>{{cite web |url=http://www.computerworld.com/action/article.do?command=viewArticleBasic&taxonomyName=hardware&articleId=9095318&taxonomyId=12&intsrc=kc_top |title=IBM's Roadrunner smashes 4-minute mile of supercomputing |access-date=June 10, 2008 |author=Sharon Gaudin |date=June 9, 2008 |publisher=Computerworld |archive-url=https://web.archive.org/web/20081224001155/http://www.computerworld.com/action/article.do?command=viewArticleBasic&taxonomyName=hardware&articleId=9095318&taxonomyId=12&intsrc=kc_top |archive-date=December 24, 2008 |df=mdy-all }}</ref><ref>{{cite web|url=http://www.top500.org/lists/2008/11/press-release |title=Austin ISC08 |publisher=Top500.org |date=November 14, 2008 |access-date=February 9, 2012 |archive-url=https://web.archive.org/web/20120222023827/http://www.top500.org/lists/2008/11/press-release |archive-date=February 22, 2012 |df=mdy }}</ref> The computer is located at Los Alamos National Laboratory in New Mexico. The computer's name refers to the New Mexico [[List of U.S. state birds|state bird]], the [[greater roadrunner]] (''Geococcyx californianus'').<ref>{{cite news|url=http://news.bbc.co.uk/1/hi/technology/7443557.stm|title=Supercomputer sets petaflop pace |last=Fildes|first=Jonathan |date=June 9, 2008|publisher=BBC News|access-date=July 8, 2008}}</ref>

In June 2008, AMD released ATI Radeon HD 4800 series, which are reported to be the first GPUs to achieve one teraFLOPS. On August 12, 2008, AMD released the ATI Radeon HD 4870X2 graphics card with two [[Radeon HD 4000 series|Radeon R770]] GPUs totaling 2.4 teraFLOPS.

In November 2008, an upgrade to the Cray [[Jaguar (supercomputer)|Jaguar supercomputer]] at the Department of Energy's (DOE's) Oak Ridge National Laboratory (ORNL) raised the system's computing power to a peak 1.64 petaFLOPS, making Jaguar the world's first petaFLOPS system dedicated to [[open research]]. In early 2009 the supercomputer was named after a mythical creature, [[Kraken]]. Kraken was declared the world's fastest university-managed supercomputer and sixth fastest overall in the 2009 TOP500 list. In 2010 Kraken was upgraded and can operate faster and is more powerful.

In 2009, the [[Cray]] Jaguar performed at 1.75 petaFLOPS, beating the IBM Roadrunner for the number one spot on the [[TOP500]] list.<ref>{{cite news| url=https://www.forbes.com/2009/11/15/supercomputer-ibm-jaguar-technology-cio-network-cray.html?feed=rss_popstories | work=Forbes | first=Andy | last=Greenberg | title=Cray Dethrones IBM in Supercomputing | date=November 16, 2009}}</ref>

In October 2010, China unveiled the [[Tianhe-1]], a supercomputer that operates at a peak computing rate of 2.5 petaFLOPS.<ref>{{cite news| url=https://www.bbc.co.uk/news/technology-11644252 | publisher=BBC News | title=China claims supercomputer crown | date=October 28, 2010}}</ref><ref>{{cite web|last=Dillow |first=Clay |url=http://www.popsci.com/technology/article/2010-10/china-unveils-2507-petaflop-supercomputer-worlds-fastest |title=China Unveils 2507 Petaflop Supercomputer, the World's Fastest |website=Popsci.com |date=October 28, 2010 |access-date=February 9, 2012 }}</ref>

{{As of|2010}} the fastest PC [[microprocessor|processor]] reached 109&nbsp;gigaFLOPS (Intel Core i7 [[Gulftown (microprocessor)|980 XE]])<ref>{{Cite web |url=http://techgage.com/article/intels_core_i7-980x_extreme_edition_-_ready_for_sick_scores/8 |title=Intel's Core i7-980X Extreme Edition – Ready for Sick Scores?: Mathematics: Sandra Arithmetic, Crypto, Microsoft Excel |website=Techgage |date=March 10, 2010 |access-date=February 9, 2012}}</ref> in double precision calculations. [[Graphics processing unit|GPU]]s are considerably more powerful. For example, [[Nvidia Tesla]] C2050 GPU computing processors perform around 515 gigaFLOPS<ref name="nvidia.com">{{cite web|url=http://www.nvidia.com/object/product_tesla_C2050_C2070_us.html |title=NVIDIA Tesla Personal Supercomputer |publisher=Nvidia.com |access-date=February 9, 2012}}</ref> in double precision calculations, and the AMD FireStream 9270 peaks at 240 gigaFLOPS.<ref name="ati.amd.com">{{cite web|url=https://www.amd.com/us/products/workstation/firestream/firestream-9270/pages/firestream-9270.aspx |title=AMD FireStream 9270 GPU Compute Accelerator |publisher=Amd.com |access-date=February 9, 2012}}</ref>

In November 2011, it was announced that Japan had achieved 10.51 petaFLOPS with its [[K computer]].<ref name="Petaflops">{{cite web|url=http://www.fujitsu.com/global/news/pr/archives/month/2011/20111102-02.html |title='K computer' Achieves Goal of 10 Petaflops |publisher=Fujitsu.com |access-date=February 9, 2012}}</ref> It has 88,128 [[SPARC64 VIIIfx]] [[central processing unit|processor]]s in 864 racks, with theoretical performance of 11.28 petaFLOPS. It is named after the Japanese word "[[wikt:京#Japanese|kei]]", which stands for 10 [[1,000,000,000,000,000|quadrillion]],<ref>See [[Japanese numerals#Large numbers|Japanese numbers]]</ref> corresponding to the target speed of 10 petaFLOPS.

On November 15, 2011, Intel demonstrated a single x86-based processor, code-named "Knights Corner", sustaining more than a teraFLOPS on a wide range of [[DGEMM]] operations. Intel emphasized during the demonstration that this was a sustained teraFLOPS (not "raw teraFLOPS" used by others to get higher but less meaningful numbers), and that it was the first general purpose processor to ever cross a teraFLOPS.<ref>{{cite web|url=http://www.tomshardware.com/news/intel-knights-corner-mic-co-processor,14002.html |title=Intel's Knights Corner: 50+ Core 22nm Co-processor|access-date=November 16, 2011|date=November 16, 2011}}</ref><ref>{{cite web |url=http://www.eetimes.com/electronics-news/4230654/Intel-unveils-1-TFLOP-s-Knight-s-Corner |title=Intel unveils 1&nbsp;TFLOP/s Knight's Corner |access-date=November 16, 2011 }}</ref>

On June 18, 2012, [[IBM Sequoia|IBM's Sequoia supercomputer system]], based at the U.S. Lawrence Livermore National Laboratory (LLNL), reached 16 petaFLOPS, setting the world record and claiming first place in the latest TOP500 list.<ref name="IBM Computer Sets Speed Record">{{cite news|last=Clark|first=Don|title=IBM Computer Sets Speed Record|url=https://www.wsj.com/articles/SB10001424052702303379204577472773983130902|access-date=June 18, 2012|newspaper=The Wall Street Journal|date=June 18, 2012}}</ref>

On November 12, 2012, the TOP500 list certified [[Titan (supercomputer)|Titan]] as the world's fastest supercomputer per the LINPACK benchmark, at 17.59 petaFLOPS.<ref>{{cite news|url=https://www.bbc.co.uk/news/technology-20272810 |title=US Titan supercomputer clocked as world's fastest |publisher=BBC |date=November 12, 2012 |access-date=February 28, 2013}}</ref><ref>{{cite web|url=http://top500.org/blog/lists/2012/11/press-release/ |title=Oak Ridge Claims No. 1 Position on Latest TOP500 List with Titan &#124; TOP500 Supercomputer Sites |publisher=Top500.org |date=November 12, 2012 |access-date=February 28, 2013}}</ref> It was developed by Cray Inc. at the [[Oak Ridge National Laboratory]] and combines AMD Opteron processors with "Kepler" NVIDIA Tesla graphics processing unit (GPU) technologies.<ref>{{cite web |last=Montalbano |first=Elizabeth |url=http://www.informationweek.com/news/government/enterprise-architecture/231900554 |title=Oak Ridge Labs Builds Fastest Supercomputer |website=Informationweek |date=October 11, 2011 |access-date=February 9, 2012}}</ref><ref>{{cite web |last=Tibken |first=Shara |url=http://news.cnet.com/8301-11386_3-57541791-76/titan-supercomputer-debuts-for-open-scientific-research/ |title=Titan supercomputer debuts for open scientific research &#124; Cutting Edge |website=News.CNet.com |date=October 29, 2012 |access-date=February 28, 2013}}</ref>

On June 10, 2013, China's [[Tianhe-2]] was ranked the world's fastest with 33.86 petaFLOPS.<ref>{{cite magazine|url=https://www.forbes.com/sites/alexknapp/2013/06/17/chinese-supercomputer-is-now-the-worlds-fastest-by-a-lot/ |title=Chinese Supercomputer Is Now The World's Fastest – By A Lot|magazine=Forbes Magazine |date=June 17, 2013 |access-date=June 17, 2013}}</ref>

On June 20, 2016, China's [[Sunway TaihuLight]] was ranked the world's fastest with 93 petaFLOPS on the LINPACK benchmark (out of 125 peak petaFLOPS). The system was installed at the National Supercomputing Center in Wuxi, and represented more performance than the next five most powerful systems on the TOP500 list did at the time combined.<ref>{{cite web|last=Feldman|first=Michael|title=China Races Ahead in TOP500 Supercomputer List, Ending US Supremacy|url=https://www.top500.org/news/china-races-ahead-in-top500-supercomputer-list-ending-us-supremacy/ |website=Top500.org |access-date=December 31, 2016 }}</ref>

In June 2019, [[Summit (supercomputer)|Summit]], an IBM-built supercomputer now running at the Department of Energy's (DOE) Oak Ridge National Laboratory (ORNL), captured the number one spot with a performance of 148.6 petaFLOPS on High Performance Linpack (HPL), the benchmark used to rank the TOP500 list. Summit has 4,356 nodes, each one equipped with two 22-core Power9 CPUs, and six NVIDIA Tesla V100 GPUs.<ref>{{Cite web |url=https://www.top500.org/lists/2018/06/ |title=June 2018 |website=Top500.org |access-date=2018-07-17 }}</ref>

In June 2022, the United States' [[Frontier (supercomputer)|Frontier]] was the most powerful supercomputer on TOP500, reaching 1102 petaFlops (1.102 exaFlops) on the LINPACK benchmarks.
<ref>{{cite web | url=https://en.wikipedia.org/wiki/TOP500 | title=TOP500 }}</ref>{{Circular reference|date=February 2025}}

In November 2024, the United States’ [[El Capitan (supercomputer)|El Capitan]] [[Exascale computing|exascale]] [[supercomputer]], hosted at the [[Lawrence Livermore National Laboratory]] in [[Livermore, California|Livermore]], displaced Frontier as the [[TOP500|world's fastest supercomputer]] in the 64th edition of the [[TOP500|Top500 (Nov 2024)]].

===Distributed computing records===
[[Distributed computing]] uses the Internet to link personal computers to achieve more FLOPS:

* {{As of|2020|4}}, the [[Folding@home]] network has over 2.3 exaFLOPS of total computing power.<ref>{{Cite web|url=https://stats.foldingathome.org/os|title=Folding@Home Active CPUs & GPUs by OS|website=foldingathome.org|access-date=2020-04-08}}</ref><ref>{{Cite web|url=https://twitter.com/foldingathome/status/1242918035788365830|title=Thanks to our AMAZING community, we've crossed the exaFLOP barrier! That's over a 1,000,000,000,000,000,000 operations per second, making us ~10x faster than the IBM Summit!pic.twitter.com/mPMnb4xdH3|last=Folding@home|date=March 25, 2020|website=@foldingathome|language=en|access-date=2020-04-04}}</ref><ref>{{Cite web|url=https://www.extremetech.com/computing/308332-foldinghome-crushes-exascale-barrier-now-faster-than-dozens-of-supercomputers|title=Folding@Home Crushes Exascale Barrier, Now Faster Than Dozens of Supercomputers - ExtremeTech|website=extremetech.com|access-date=2020-04-04}}</ref><ref>{{Cite web|url=https://www.techspot.com/news/84561-foldinghome-exceeds-15-exaflops-battle-against-covid-19.html|title=Folding@Home exceeds 1.5 ExaFLOPS in the battle against Covid-19|website=TechSpot|date=March 26, 2020 |language=en-US|access-date=2020-04-04}}</ref> It is the most powerful distributed computer network, being the first ever to break 1 exaFLOPS of total computing power. This level of performance is primarily enabled by the cumulative effort of a vast array of powerful [[Graphics processing unit|GPU]] and [[Central processing unit|CPU]] units.<ref>{{cite press release
 |title        = Sony Computer Entertainment's Support for Folding@home Project on PlayStation™3 Receives This Year's "Good Design Gold Award"
 |url          = http://www.scei.co.jp/corporate/release/081106de.html
 |publisher    = Sony Computer Entertainment Inc.
 |date         = November 6, 2008
 |access-date  = December 11, 2008
 |url-status   = dead
 |archive-url  = https://web.archive.org/web/20090131082202/http://www.scei.co.jp/corporate/release/081106de.html
 |archive-date = January 31, 2009
 |df           = mdy
}}</ref>

* {{As of|2020|012}}, the entire [[BOINC]] network averages about 31&nbsp;petaFLOPS.<ref>{{cite web |url=http://boinc.berkeley.edu/computing.php |title=BOINC Computing Power |publisher=BOINC |access-date=December 28, 2020}}</ref>
* {{As of|2018|06}}, [[SETI@home]], employing the [[BOINC]] software platform, averages 896&nbsp;teraFLOPS.<ref>{{cite web |url=http://boincstats.com/en/stats/0/project/detail |title=SETI@Home Credit overview |publisher=BOINC |access-date=June 15, 2018}}</ref>
* {{As of|2018|06}}, [[Einstein@Home]], a project using the [[BOINC]] network, is crunching at 3&nbsp;petaFLOPS.<ref>{{cite web |url=http://boincstats.com/en/stats/5/project/detail |title=Einstein@Home Credit overview |publisher=BOINC |access-date=June 15, 2018}}</ref>
* {{As of|2018|06}}, [[MilkyWay@home]], using the [[BOINC]] infrastructure, computes at 847&nbsp;teraFLOPS.<ref>{{cite web|url=http://boincstats.com/en/stats/61/project/detail|title=MilkyWay@Home Credit overview|publisher=BOINC|access-date=June 15, 2018}}</ref>
* {{As of|2020|06}}, [[Great Internet Mersenne Prime Search|GIMPS]], searching for [[Mersenne prime]]s, is sustaining 1,354&nbsp;teraFLOPS.<ref>{{cite web |url=http://www.mersenne.org/primenet |title=Internet PrimeNet Server Distributed Computing Technology for the Great Internet Mersenne Prime Search |work=GIMPS |access-date=June 15, 2018 }}</ref>

==Cost of computing==
===Hardware costs===
{| class="wikitable"
|-
! rowspan=2 | Date
! colspan=2 | Approximate USD per GFLOPS
! rowspan=2 | Platform providing the lowest cost per GFLOPS 
! rowspan=2 | Comments
|-
! Unadjusted
! {{Inflation-year|US}}{{Inflation-fn|US}}
|-
|1945
|$1.265[[trillion|T]]
|${{Inflation|US|1.265|1945|r=3|fmt=c}}T
|[[ENIAC]]: {{US$|long=no|487000}} in 1945 and ${{Inflation|US|487000|1945|fmt=c|r=-3}} in 2023.
|{{US$|long=no|487000}} / {{val|0.000000385|ul=GFLOPS}}. [[Vacuum-tube computer|First-generation]] ([[vacuum tube]]-based) electronic digital computer.
|-
| 1961
| $18.672[[billion|B]]
| ${{Inflation|US|18.672|1961|r=3|fmt=c}}B
| A basic installation of [[IBM 7030 Stretch]] had a cost at the time of {{US$|7.78 million}} each.
| The [[IBM 7030 Stretch]] performs one floating-point multiply every {{val|2.4 |ul=microseconds}}.<ref>{{cite web|url=http://computer-history.info/Page4.dir/pages/IBM.7030.Stretch.dir/ |title=The IBM 7030 (STRETCH) |publisher=Norman Hardy |access-date=February 24, 2017}}</ref> [[Transistor computer|Second-generation]] (discrete [[transistor]]-based) computer.
|-
| 1964
| $2.3B
| ${{Inflation|US|2.3|1964|r=3|fmt=c}}B
| Base model [[CDC 6600]] price: $6,891,300.
| The CDC 6600 is considered to be the first commercially-successful [[supercomputer]].
|-
| 1984
| $18,750,000
| ${{Inflation|US|18750000|1984|r=0|fmt=c}}
| [[Cray X-MP]]/48
| $15,000,000 / 0.8 GFLOPS. Third-generation ([[integrated circuit]]-based) computer.
|-
| 1997
| $30,000
| ${{Inflation|US|30000|1997|r=0|fmt=c}}
| Two 16-processor [[Beowulf (computing)|Beowulf]] clusters with [[Pentium Pro]] microprocessors<ref>{{cite web |url=http://loki-www.lanl.gov/papers/sc97/ |title=Loki and Hyglac |publisher=Loki-www.lanl.gov |date=July 13, 1997 |access-date=February 9, 2012 |archive-url=https://web.archive.org/web/20110721043504/http://loki-www.lanl.gov/papers/sc97/ |archive-date=July 21, 2011 |url-status=dead }}</ref>
|
|-
| {{sort|2000/04|April 2000}}
| $1,000
| ${{Inflation|US|1016|2000|r=0|fmt=c}} <!-- $1000 in April 2000 had value $1016 in December 2000 using https://data.bls.gov/cgi-bin/cpicalc.pl, Inflation template only accepts years (end-of-year) -->
| [[Beowulf cluster|Bunyip Beowulf cluster]] <!-- old link bad as of 2013-05-18 http://tsg.anu.edu.au/Projects/Beowulf/ -->
| Bunyip was the first sub-{{val|p={{US$}}|1|upl=MFLOPS}} computing technology. It won the Gordon Bell Prize in 2000.
|-
| {{sort|2000/05|May 2000}}
| $640
| ${{Inflation|US|640|2000|r=0|fmt=c}} <!-- $640 in May 2000 had value $649 in December 2000 using https://data.bls.gov/cgi-bin/cpicalc.pl, Inflation template only accepts years (end-of-year) -->
| [[Kentucky Linux Athlon Testbed|KLAT2]]
| KLAT2 was the first computing technology which scaled to large applications while staying under {{val|p={{US$}}|1|up=MFLOPS}}.<ref>{{Cite web |url=http://aggregate.org/KLAT2/ |title=Kentucky Linux Athlon Testbed 2 (KLAT2) |website=The Aggregate |access-date=February 9, 2012}}</ref>
|-
| {{sort|2003/08|August 2003}}
| $83.86
| ${{Inflation|US|83.86|2003|r=2|fmt=c}}
| KASY0
| KASY0 was the first sub-{{val|p={{US$}}100|up=GFLOPS}} computing technology. KASY0 achieved 471 GFLOPS on 32-bit HPL. At a cost of less than $39,500, that makes it the first supercomputer to break $100/GFLOPS.<ref>{{Cite web |url=https://www.haveland.com/index.htm?povbench/index.php |title=Haveland-Robinson Associates - Home Page|website=Haveland-Robinson Associates |date=August 23, 2003 |access-date=November 14, 2024}}</ref>
|-
| {{sort|2007/08|August 2007}}
| $48.31
| ${{Inflation|US|48.31|2007|r=2|fmt=c}}
| Microwulf
| As of August 2007, this {{val|26|u=GFLOPS}} "personal" Beowulf cluster can be built for $1256.<ref>{{cite web|url=http://www.calvin.edu/~adams/research/microwulf/ |title=Microwulf: A Personal, Portable Beowulf Cluster |access-date=February 9, 2012 |url-status=dead |archive-url=https://web.archive.org/web/20070912061302/http://www.calvin.edu/~adams/research/microwulf/ |archive-date=September 12, 2007 }}</ref>
|-
| {{sort|2011/03|March 2011}}
| $1.80
| ${{Inflation|US|1.80|2011|r=2|fmt=c}} <!-- $1.80 in March 2011 had value $1.82 in December 2011 using https://data.bls.gov/cgi-bin/cpicalc.pl, Inflation template only accepts years (end-of-year) -->
| HPU4Science
| This $30,000 cluster was built using only commercially available "gamer" grade hardware.<ref>Adam Stevenson, Yann Le Du, and Mariem El Afrit. "[https://arstechnica.com/science/news/2011/03/high-performance-computing-on-gamer-pcs-part-1-hardware.ars High-performance computing on gamer PCs]." ''Ars Technica''. March 31, 2011.</ref>
|-
| {{sort|2012/08|August 2012}}
| 75¢
| ${{Inflation|US|.75|2012|r=2|fmt=c}}
| Quad [[Radeon HD 7000 series|AMD Radeon 7970]] System
| A quad [[AMD]] [[Radeon HD 7000 series|Radeon 7970]] desktop computer reaching 16 TFLOPS of single-precision, 4 TFLOPS of double-precision computing performance. Total system cost was $3000; built using only commercially available hardware.<ref>{{cite web |url=http://www.overclock3d.net/reviews/gpu_displays/hd7970_quadfire_eyefinity_review/12 |title=HD7970 Quadfire Eyefinity Review |date=January 9, 2012 |website=OC3D.net |author=Tom Logan}}</ref>
|-
| {{sort|2013/06|June 2013}}
| 21.68¢
| {{Inflation|US|21.68|2013|r=2|fmt=c}}¢
| [[PlayStation 4|Sony PlayStation 4]]
| The Sony [[PlayStation 4]] is listed as having a peak performance of {{val|1.84|ul=TFLOPS}}, at a price of $399<ref>"[https://www.cnbc.com/id/100805004 Sony Sparks Price War With PS4 Priced at $399]." ''CNBC''. June 11, 2013.</ref>
|-
| {{sort|2013/11|November 2013}}
| 16.11¢
| {{Inflation|US|16.11|2013|r=2|fmt=c}}¢
| [[Sempron|AMD Sempron 145]] & [[GeForce 700 series|GeForce GTX 760]] system
| Built using commercially available parts, a system using one AMD [[Sempron]] 145 and three [[Nvidia]] [[GeForce 700 series|GeForce GTX 760]] reaches a total of {{val|6.771|u=TFLOPS}} for a total cost of {{US$|1090.66}}.<ref>{{Cite web | url=http://pcpartpicker.com/p/22JOc | archive-url=http://www.freezepage.com/1384601420XCIGYKCBKJ?url=http://pcpartpicker.com/p/22JOc | url-status=dead | archive-date=November 16, 2013 | title=FreezePage | access-date=May 9, 2020 }}</ref>
|-
| {{sort|2013/12|December 2013}}
| 12.41¢
| {{Inflation|US|12.41|2013|r=2|fmt=c}}¢
| [[Sandy Bridge|Pentium G550]] & [[Radeon Rx 200 series|Radeon R9 290]] system
| Built using commercially available parts. [[Intel]] [[Sandy Bridge|Pentium G550]] and AMD [[AMD Radeon Rx 200 series|Radeon R9 290]] tops out at {{val|4.848|u=TFLOPS}} grand total of {{US$|681.84}}.<ref>{{Cite web | url=http://pcpartpicker.com/p/2mQxd | archive-url=http://www.freezepage.com/1387480124PSLSILVCMJ?url=http://pcpartpicker.com/p/2mQxd | url-status=dead | archive-date=December 19, 2013 | title=FreezePage | access-date=May 9, 2020 }}</ref>
|-
| {{sort|2015/01|January 2015}}
| 7.85¢
| {{Inflation|US|7.85|2015|r=2|fmt=c}}¢
| [[Haswell (microarchitecture)|Celeron G1830]] & [[Radeon Rx 200 series|Radeon R9 295X2]] system
| Built using commercially available parts. Intel [[Haswell (microarchitecture)|Celeron G1830]] and AMD [[AMD Radeon Rx 200 series|Radeon R9 295X2]] tops out at over {{val|11.5|u=TFLOPS}} at a grand total of {{US$|902.57}}.<ref>{{Cite web | url=http://pcpartpicker.com/p/8z3cVn | archive-url=http://www.freezepage.com/1420850340WGSMHXRBLE?url=http://pcpartpicker.com/p/8z3cVn | url-status=dead | archive-date=January 10, 2015 | title=FreezePage | access-date=May 9, 2020 }}</ref><ref>{{Cite web | url=http://www.tomshardware.com/reviews/radeon-r9-295x2-review-benchmark-performance,3799.html | title=Radeon R9 295X2 8 GB Review: Project Hydra Gets Liquid Cooling| date=April 8, 2014}}</ref>
|-
| {{sort|2017/07|June 2017}}
| 6¢
| {{Inflation|US|6.00|2017|r=2|fmt=c}}¢
| [[Zen (first generation)|AMD Ryzen 7 1700]] & [[Radeon Pro|AMD Radeon Vega Frontier Edition]] system
| Built using commercially available parts. AMD Ryzen 7 1700 CPU combined with AMD Radeon Vega FE cards in CrossFire tops out at over {{val|50|u=TFLOPS}} at just under {{USD|3,000}} for the complete system.<ref>{{cite web|url=https://medium.com/intuitionmachine/building-a-50-teraflops-amd-vega-deep-learning-box-for-under-3k-ebdd60d4a93c|title=Building a 50 Teraflops AMD Vega Deep Learning Box for Under $3K|last=Perez|first=Carol E.|date=July 13, 2017|work=Intuition Machine|access-date=July 26, 2017}}</ref>
|-
|October 2017
|2.73¢
|{{Inflation|US|2.73|2017|r=2|fmt=c}}¢
|[[Kaby Lake|Intel Celeron G3930]] & [[Radeon RX Vega series|AMD RX Vega 64]] system
|Built using commercially available parts. Three [[AMD RX Vega series|AMD RX Vega 64]] graphics cards provide just over 75 TFLOPS half precision (38 TFLOPS SP or 2.6 TFLOPS DP when combined with the CPU) at ~$2,050 for the complete system.<ref>{{Cite web|url=https://pcpartpicker.com/user/mattebaughman/saved/8DQZ8d|title=lowest_$/fp16 - mattebaughman's Saved Part List - Celeron G3930 2.9GHz Dual-Core, Radeon RX VEGA 64 8GB (3-Way CrossFire), XON-350_BK ATX Mid Tower|website=pcpartpicker.com|access-date=2017-09-13}}</ref>
|-
|November 2020
|3.14¢
|{{Inflation|US|3.14|2020|r=2|fmt=c}}¢
|[[Zen 2|AMD Ryzen 3600]] & 3× [[GeForce 30 series|NVIDIA RTX 3080]] system
|AMD Ryzen 3600 @ 484 GFLOPS & $199.99

3× NVIDIA RTX 3080 @ 29,770 GFLOPS each & $699.99

Total system GFLOPS = 89,794 / TFLOPS = 89.794

Total system cost incl. realistic but low cost parts; matched with other example = $2839<ref>{{Cite web|title=System Builder|url=https://pcpartpicker.com/list/9bPn8J|access-date=2020-12-07|website=pcpartpicker.com}}</ref>

{{US$}}/GFLOP = $0.0314
|-
|November 2020
|3.88¢
|{{Inflation|US|3.88|2020|r=2|fmt=c}}¢
|[[PlayStation 5]]
|The Sony [[PlayStation 5]] Digital Edition is listed as having a peak performance of 10.28 TFLOPS (20.56 TFLOPS at half precision) at a retail price of $399.<ref>{{cite web |url=https://www.techpowerup.com/gpu-specs/playstation-5-gpu.c3480 |title=AMD Playstation 5 GPU Specs |website=techpowerup.com |access-date=May 12, 2021}}</ref>
|-
|November 2020
|4.11¢
|{{Inflation|US|4.11|2020|r=2|fmt=c}}¢
|[[Xbox Series X and Series S|Xbox Series X]]
|Microsoft's [[Xbox Series X]] is listed as having a peak performance of 12.15 TFLOPS (24.30 TFLOPS at half precision) at a retail price of $499.<ref>{{cite web |url=https://www.xbox.com/en-US/consoles/xbox-series-x?xr=shellnav#specs |title=Xbox Series X &#124; Xbox |website=xbox.com |access-date=September 21, 2021}}</ref>
|-
|September 2022
|1.94¢
|{{Inflation|US|1.94|2022|r=2|fmt=c}}¢
|[[RTX 4090]]
|Nvidia's [[RTX 4090]] is listed as having a peak performance of 82.6 TFLOPS (1.32 PFLOPS at 8-bit precision) at a retail price of $1599.<ref>{{cite web |url=https://www.tomshardware.com/news/nvidia-geforce-rtx-4090-rtx-4080-price-release-date-specs-revealed |title=Nvidia Announces RTX 4090 Coming October 12, RTX 4080 Later |website=tomshardware.com |date=September 20, 2022 |access-date=September 20, 2022}}</ref>
|-
|May 2023
|1.25¢
|{{Inflation|US|1.25|2023|r=2|fmt=c}}¢
|[[Radeon RX 7000 series|Radeon RX 7600]]
|AMD's [[Radeon RX 7000 series|RX 7600]] is listed as having a peak performance of 21.5 TFLOPS at a retail price of $269.<ref>{{cite web |url=https://www.tomshardware.com/reviews/amd-radeon-rx-7600-review |title=AMD Radeon RX 7600 Review: Incremental Upgrades |website=tomshardware.com |date=May 24, 2023 |access-date=May 24, 2023}}</ref>
|-
|}

<!---
START OF A BUNCH OF DELETED STUFF
--->
<!-- These were removed because they only cited the price of the processor; this table is supposed to compare the total cost of the machine. -----------
|-
|Q4 2018
|$0.02
|${{Inflation|US|0.02|2018|r=2|fmt=c}}
|$20
|[[Nvidia Jetson|Nvidia Jetson AGX Xavier]]
|The ''Jetson AGX Xavier Developer Kit'' provides more than 30 TOPS (trillion operations per second) combined for [[deep learning]] and [[computer vision]] tasks at $699 and 30 Watts: a 512-core Volta GPU with Tensor Cores gives up to 11 TFLOPS FP16 or 22 TOPS INT8 compute, dual NVDLA engines give 10 TOPS INT8 or 5 TFLOPS FP16. It also has high-performance eight-core ARM64 CPU with additional dedicated co-processors for accelerating computer vision tasks: one image processor, a video processor and a vision processor.<ref name="NVIDIA_(2018)">{{cite web |title=Jetson FAQ |url=https://developer.nvidia.com/embedded/faq#xavier-performance |website=NVIDIA Autonomous Machines |publisher=NVIDIA Corporation |access-date=21 October 2019}}</ref><ref>[https://developer.nvidia.com/embedded/buy/jetson-xavier-devkit NVIDIA Jetson Xavier Developer Kit]</ref><ref name="LarabelM_(2018)">{{cite web |last1=Larabel |first1=Michael |title=NVIDIA Jetson AGX Xavier Benchmarks - Incredible Performance On The Edge |url=https://www.phoronix.com/scan.php?page=article&item=nvidia-jetson-xavier&num=1 |website=phoronix |publisher=Phoronix Media 2018 |access-date=21 October 2019}}</ref>
|-
| {{sort|2007, March|March 2007}}
| $0.42
| [[Ambric]] AM2045<ref>{{cite journal|last=Halfill|first=Tom R.|date=October 10, 2006|title=Ambric's New Parallel Processor|journal=Microprocessor Report|publisher=Reed Electronics Group|pages=1–9|url=http://www.ambric.com/pdf/MPR_Ambric_Article_10-06_204101.pdf|access-date=July 8, 2008|archive-url = https://web.archive.org/web/20080627111128/http://www.ambric.com/pdf/MPR_Ambric_Article_10-06_204101.pdf |archive-date = June 27, 2008|url-status=dead}}</ref>
|
|-
| {{sort|2009, September|September 2009}}
| $0.13 (single precision)
| [[ATI Technologies|ATI]] [[Evergreen (GPU family)|Radeon R800]]<ref>{{cite news|url=http://www.brightsideofnews.com/news/2009/9/29/the-fastest-ati-5870-card-achieves-3tflops!.aspx|title=The fastest ATI 5870 card achieves 3TFLOPS!|author=Valich, Theo|date=September 29, 2009|publisher=Bright Side of News|access-date=September 29, 2009}}</ref>
|The first high-performance 40&nbsp;nm [[Graphics processing unit|GPU]] from ATI. It can reach speeds of 3.04&nbsp;TFLOPS when running at 950&nbsp;MHz. Price per GFLOPS is slightly inaccurate as it is single precision and includes only the cost of the card.
|-
| {{sort|2011, March|March 2011}}
| $0.13
| [[Advanced Micro Devices|AMD]] [[Comparison of AMD graphics processing units#Northern Islands .28HD 6xxx.29 series|Radeon HD 6990]] [[Overclocking|Overclocked]]<ref>http://www.hardocp.com/images/articles/1299536835FpEmksdSXb_1_3_l.gif {{Dead link|date=February 2022}}</ref>
|Floating-point performance (peak): 5.40&nbsp;TFLOPS. Price: $699.
|}
|-
| {{sort|2013/04|April 2013}}
| $0.12
| $0.12
| [https://www.amd.com/us/products/desktop/graphics/7000/7990/pages/radeon-7990.aspx#2 AMD Radeon HD 7990]
| The AMD Radeon HD 7990 is a GPU with single precision computing performance reaching 8.2&nbsp;TFLOPS. It was released on April 24, 2013 with a price point of $1000.
---->
<!--* 2006, February: about $1 per GFLOPS in ATI PC add-in graphics card (X1900 architecture) – these figures are disputed as they refer to highly parallelized GPU power-->
<!-- The source does not state whether FLOPS is SP or DP, which is misleading: * 2007, October: about $0.20 per GFLOPS with the cheapest retail [[PlayStation 3|Sony PS3]] console, at US$400, that runs at a claimed 2 teraFLOPS; these figures represent the processing power of the [[Graphics processing unit|GPU]]. The seven [[Central processing unit|CPU]]s run collectively at a lower 218&nbsp;GFLOPS.<ref>{{cite news|url=http://news.bbc.co.uk/2/hi/technology/4554025.stm|title=Sony shows off new PlayStation 3|last=Hermida |first=Alfred |date=May 17, 2005|work=BBC News|access-date=July 8, 2008}}</ref> --><!--* 2008, June/July: ~20c (€) per GFLOPS with AMD's HD4870 GPU. These figures are disputed as they refer to highly parallelized GPU power-->
<!---
END OF A BUNCH OF DELETED STUFF
--->

==See also==
{{div col|colwidth=20em}}
* [[Computer performance by orders of magnitude]]
* [[Exascale computing]]
* [[Gordon Bell Prize]]
* [[LINPACK benchmarks]]
* [[Moore's law]]
* [[Multiply–accumulate operation]]
* [[Performance per watt#FLOPS per watt]]
* [[SPECfp]]
* [[SPECint]]
* [[SUPS]]
* [[TOP500]]
{{div col end}}
{{clear}}

==References==
{{Reflist|refs=
}}

{{Graphics Processing Unit}}
{{CPU technologies}}

{{Authority control}}

{{DEFAULTSORT:Flops}}
[[Category:Benchmarks (computing)]]
[[Category:Floating point]]
[[Category:Units of frequency]]