View source for FLOPS

{{Other uses|Flop (disambiguation){{!}}Flop}}
{{Use mdy dates|date=May 2015}}
{{More citations needed|date=February 2015|reason=introduction is argued, but not sourced}}
{| class="infobox"
|+ Computer performance
! 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>
|-
|}

In [[computing]], '''floating point operations per second''' ('''FLOPS''', '''flops''' or '''flop/s''') is a measure of [[computer performance]], useful in fields of scientific computations that require [[floating-point]] calculations. For such cases it is a more accurate measure than measuring [[instructions per second]].

The similar term FLOP is often used for ''floating-point operation'', for example as a unit of counting floating-point operations carried out by an algorithm or computer hardware.

==Floating-point arithmetic==
[[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 everything is carried out in base two, rather than 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 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] 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. The unit MIPS measures integer performance of a computer. Examples of integer operation include data movement (A to B) or value testing (If A = B, then C). 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> Frank H. McMahon, of the Lawrence Livermore National Laboratory, invented the terms FLOPS and MFLOPS (megaFLOPS) so that he could compare the supercomputers of the day by the number of floating-point calculations they performed per second. 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.

FLOPS can be calculated using this equation:<ref name="en.community.dell.com">[https://www.dell.com/support/article/fr/fr/frbsdt1/sln310893/nodes-sockets-cores-and-flops-oh-my "Nodes, Sockets, Cores and FLOPS, Oh, My" by Dr. Mark R. Fernandez, Ph.D.]</ref>

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

==FLOPs per cycle for various processors==
{| class="wikitable sortable"
! scope="col" | Microarchitecture
! scope="col" | ISA
! scope="col" | FP64
! scope="col" | FP32
! scope="col" | FP16
|-

| Intel [[Atom (system on chip)|Atom]] (Bonnell, Saltwell, Silvermont and Goldmont) || [[SSE3]] (64-bit) || 2 || 4 || 0
|-
| Intel [[Intel Core (microarchitecture)|Core]] (Merom, [[Penryn (microarchitecture)|Penryn]])<br />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 }}</ref> ([[Nehalem (microarchitecture)|Nehalem]], [[Westmere (microarchitecture)|Westmere]]) || [[SSE4]] (128-bit) || 4 || 8 || 0
|-
| Intel [[Sandy Bridge]] ([[Sandy Bridge]], [[Ivy Bridge (microarchitecture)|Ivy Bridge]]) || [[Advanced Vector Extensions|AVX]] (256-bit) || 8 || 16 || 0
|-
| Intel [[Haswell (microarchitecture)|Haswell]]<ref name="tpeak_jos"/> ([[Haswell (microarchitecture)|Haswell]], [[Haswell (microarchitecture)|Devil's Canyon]], [[Broadwell (microarchitecture)|Broadwell]])<br />Intel [[Skylake (microarchitecture)|Skylake]] ([[Skylake (microarchitecture)|Skylake]], [[Kaby Lake]], [[Coffee Lake]], [[Whiskey Lake (microarchitecture)|Whiskey lake]], [[Amber Lake (microarchitecture)|Amber lake]]) || [[Advanced Vector Extensions|AVX2]] & FMA (256-bit) || 16 || 32 || 0
|-
| Intel [[Xeon Phi]] (Knights Corner) || [[Streaming SIMD Extensions|SSE]] & FMA (256-bit) || 16 || 32 || 0
|-
| Intel [[Skylake (microarchitecture)|Skylake-X]]<br />Intel [[Xeon Phi]] (Knights Landing, Knights Mill) || [[Advanced Vector Extensions|AVX-512]] & FMA (512-bit) || 32 || 64 || 0
|-
| AMD [[Bobcat (microarchitecture)|Bobcat]] || [[x86-64|AMD64]] (64-bit) || 2 || 4 || 0
|-
| AMD [[Jaguar (microarchitecture)|Jaguar]]<br />AMD [[Puma (microarchitecture)|Puma]]<br />AMD [[AMD 10h|K10]]<br />AMD [[Bulldozer (microarchitecture)|Bulldozer]]<ref name="tpeak_jos"/> ([[Piledriver (microarchitecture)|Piledriver]], [[Steamroller (microarchitecture)|Steamroller]], [[Excavator (microarchitecture)|Excavator]]) || [[Advanced Vector Extensions|AVX]] (128-bit) || 4 || 8 || 0
|-
| AMD [[Zen (microarchitecture)|Zen]] (Ryzen 1000 series, Threadripper 1000 series, Epyc [[Epyc|Naples]])<br />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}} [https://images.anandtech.com/doci/10591/HC28.AMD.Mike%20Clark.final-page-007.jpg page 7]</ref> (Ryzen 2000 series, Threadripper 2000 series) || [[Advanced Vector Extensions|AVX2]] & FMA (128-bit, 256-bit decoding)<ref>{{Cite web | url=https://www.agner.org/optimize/microarchitecture.pdf | title=The microarchitecture of Intel and AMD CPUs}}</ref> || 8 || 16 || 0
|-
| 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|website=www.youtube.com}}</ref> (Ryzen 3000 series, Epyc [[Epyc|Rome]]) || [[Advanced Vector Extensions|AVX2]] & FMA (256-bit) || 16 || 32 || 0
|-
| ARM Cortex-A7, A9, A15 || [[ARM architecture|ARMv7]] || 1 || 8 || 0
|-
| ARM Cortex-A32, A35, A53, A55, [[ARM Cortex-A72|A72]], [[ARM Cortex-A73|A73]], [[ARM Cortex-A75|A75]] || [[ARM architecture|ARMv8]] || 2 || 8 || 0
|-
| ARM Cortex-[[ARM Cortex-A57|A57]]<ref name="tpeak_jos"/> || [[ARM architecture|ARMv8]] || 4 || 8 || 0
|-
| ARM Cortex-[[ARM Cortex-A76|A76]], [[ARM Cortex-A77|A77]] || [[ARM architecture|ARMv8]] || 8 || 16 || 0
|-
| Qualcomm [[Krait (CPU)|Krait]] || [[ARM architecture|ARMv8]] || 1 || 8 || 0
|-
| Qualcomm [[Kryo]] (1xx - 3xx) || [[ARM architecture|ARMv8]] || 2 || 8 || 0
|-
| Qualcomm [[Kryo]] (4xx) || [[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) || ? || 8 || 8 (as FP64) || 0
|-
| Hitachi [[Hitachi SH-4|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 |accessdate=21 June 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 |accessdate=21 June 2019}}</ref> || [[Hitachi SH-4|SH-4]] || 1 || 7 || 0
|-
| Nvidia [[Fermi (microarchitecture)|Fermi]] (only GeForce GTX 465-480, 560 Ti, 570-590) || [[Parallel Thread Execution|PTX]] || 1/4 (locked by driver, 1 in hardware) || 2 || 0
|-
| Nvidia [[Fermi (microarchitecture)|Fermi]] (only Quadro 600-2000) || [[Parallel Thread Execution|PTX]] || 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]] || 1/12 (for [[GeForce 700 series|GK110]]: locked by driver, 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
|-
| Nvidia [[Maxwell (microarchitecture)|Maxwell]]<br />Nvidia [[Pascal (microarchitecture)|Pascal]] (all except Quadro GP100 and Tesla P100) || [[Parallel Thread Execution|PTX]] || 1/16 || 2 || 1/32
|-
| Nvidia [[Pascal (microarchitecture)|Pascal]] (only Quadro GP100 and Tesla P100) || [[Parallel Thread Execution|PTX]] || 1 || 2 || 4
|-
| Nvidia [[Volta (microarchitecture)|Volta]] || [[Parallel Thread Execution|PTX]] || 1 || 2 || 16
|-
| Nvidia [[Turing (microarchitecture)|Turing]] (only GeForce [[GeForce 16 series|16XX]]) || [[Parallel Thread Execution|PTX]] || 1/16 || 2 (FP32) + 2 (INT32) || 4
|-
| Nvidia [[Turing (microarchitecture)|Turing]] (all except GeForce [[GeForce 16 series|16XX]]) || [[Parallel Thread Execution|PTX]] || 1/16 || 2 (FP32) + 2 (INT32) || 16
|-
| AMD [[Graphics Core Next|GCN]] (only Radeon Pro WX 2100-7100) || [[Graphics Core Next|GCN]] || 1/8 || 2 || 2
|-
| AMD [[Graphics Core Next|GCN]] (all except Radeon VII, Instinct MI50 and MI60, Radeon Pro WX 2100-7100) || [[Graphics Core Next|GCN]] || 1/8 || 2 || 4
|-
| AMD [[AMD RX Vega series|GCN Vega 20]] (only Radeon VII) || [[Graphics Core Next|GCN]] || 1/2 (locked by driver, 1 in hardware) || 2 || 4
|-
| AMD [[AMD RX Vega series|GCN Vega 20]] (only Radeon Instinct MI50 and MI60) || [[Graphics Core Next|GCN]] || 1 || 2 || 4
|-
| 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?start=1|title=Alles zu Navi: Radeon RX 5700 XT ist RDNA mit GDDR6}}</ref><ref name="techpowerup">{{cite web|url=https://www.techpowerup.com/gpu-specs/radeon-rx-5700-xt.c3339|title=AMD Radeon RX 5700 XT}}</ref> || [[AMD RDNA Architecture|RDNA]]  || 1/8 || 2 || 4
|-
| Graphcore Colossus GC2<ref name="Source 1">{{cite web|url=https://www.youtube.com/watch?v=2IOyQEIlN6Y&t=986|title=6 threads per core imply that IPC is a multiple of 6, 1216 cores per chip|website=www.youtube.com}}</ref><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=www.youtube.com}}</ref><ref name="Source 3">{{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=www.youtube.com}}</ref> (values estimated) || ? || 0 || 18 || 72

|}<ref>{{Cite web | url=https://en.wikichip.org/wiki/flops | title=Floating-Point Operations Per Second (FLOPS)}}</ref>

==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 |accessdate=November 17, 2011 |url-status=dead |archiveurl=https://web.archive.org/web/20101105131112/http://www.jacobsequity.com/ASCI%20Red%20Supercomputer.pdf |archivedate=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.

For comparison, a [[calculator|handheld calculator]] performs relatively few FLOPS. A computer response time below 0.1&nbsp;second in a calculation context is usually perceived as instantaneous by a human operator,<ref>{{cite web |url=http://www.useit.com/papers/responsetime.html |title=Response Times: The Three Important Limits |accessdate=June 11, 2008 |publisher=Jakob Nielsen}}</ref> so a simple calculator needs only about 10&nbsp;FLOPS to be considered functional.

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 |accessdate=February 9, 2012}}</ref>

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. 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|accessdate=July 8, 2008}}</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 |publisher=Top500.org |accessdate=July 8, 2008 |url-status=dead |archiveurl=https://web.archive.org/web/20080509064814/http://www.top500.org/news/2007/06/23/29th_top500_list_world_s_fastest_supercomputers_released |archivedate=May 9, 2008 |df=mdy }}</ref> The [[Cray XT4]] hit second place with 101.7&nbsp;teraFLOPS.

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|accessdate=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 [[Texas Advanced Computing Center#Ranger|Ranger]],<ref>{{cite web
 |url         = http://www.tacc.utexas.edu/resources/hpcsystems/
 |title       = University of Texas at Austin, Texas Advanced Computing Center
 |accessdate  = September 13, 2010
 |quote       = Any researcher at a U.S. institution can submit a proposal to request an allocation of cycles on the system.
 |url-status     = dead
 |archiveurl  = https://web.archive.org/web/20090801102108/http://www.tacc.utexas.edu/resources/hpcsystems/
 |archivedate = 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 |accessdate=June 10, 2008 |author=Sharon Gaudin |date=June 9, 2008 |publisher=Computerworld |url-status=dead |archiveurl=https://web.archive.org/web/20081224001155/http://www.computerworld.com/action/article.do?command=viewArticleBasic&taxonomyName=hardware&articleId=9095318&taxonomyId=12&intsrc=kc_top |archivedate=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 |accessdate=February 9, 2012 |url-status=dead |archiveurl=https://web.archive.org/web/20120222023827/http://www.top500.org/lists/2008/11/press-release |archivedate=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|work=BBC News|accessdate=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 XT 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 | work=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 |publisher=Popsci.com |date=October 28, 2010 |accessdate=February 9, 2012}}</ref>

{{As of|2010}} the fastest six-core PC [[microprocessor|processor]] reaches 109&nbsp;gigaFLOPS ([[Intel Core#Core i7|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 |accessdate=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 |date= |accessdate=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 |date= |accessdate=February 9, 2012}}</ref> In single precision performance, Nvidia Tesla C2050 computing processors perform around 1.03 teraFLOPS and the AMD FireStream 9270 cards peak at 1.2 teraFLOPS. Both Nvidia and AMD's consumer gaming GPUs may reach higher FLOPS. For example, AMD's HemlockXT 5970<ref>{{Cite web | url=https://www.amd.com/us/products/desktop/graphics/ati-radeon-hd-5000/hd-5970/Pages/ati-radeon-hd-5970-specifications.aspx | title=Welcome to AMD &#124; Processors &#124; Graphics and Technology &#124; AMD}}</ref> reaches 928 gigaFLOPS in double precision calculations with two GPUs on board and the Nvidia GTX 480 reaches 672 gigaFLOPS<ref>{{cite web|url=http://www.nvidia.com/object/product_geforce_gtx_480_us.html |title=GeForce GTX 480 |publisher=Nvidia.com |date=July 20, 2010 |accessdate=February 9, 2012}}</ref> with one GPU on board.

On December 2, 2010, the US Air Force unveiled a defense supercomputer made up of 1,760 [[PlayStation 3]] consoles that can run 500 teraFLOPS.<ref>{{cite web|last=Dillow |first=Clay |url=http://www.popsci.com/technology/article/2010-12/air-forces-new-supercomputer-made-1760-playstation-3s |title=Air Force Unveils Fastest Defense Supercomputer, Made of 1760 PlayStation 3 |publisher=Popsci.com |date= |accessdate=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 |date= |accessdate=February 9, 2012}}</ref> It is still under development and software performance tuning is currently underway. 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:?#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|accessdate=November 16, 2011|date=2011-11-16}}</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|accessdate=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|accessdate=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=BBC News – US Titan supercomputer clocked as world's fastest |publisher=Bbc.co.uk |date=November 12, 2012 |accessdate=February 28, 2013|work=BBC News }}</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 |accessdate=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 graphic 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 |accessdate=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 – CNET News |publisher=News.cnet.com |date=October 29, 2012 |accessdate=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 |accessdate=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, which is almost exclusively based on technology developed in China, is installed at the National Supercomputing Center in Wuxi, and represents more performance than the next five most powerful systems on the TOP500 list combined.<ref>{{cite web|last1=Feldman|first1=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|accessdate=31 December 2016}}</ref>

In June 2018, [[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 '''122.3 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 {{!}} TOP500 Supercomputer Sites|website=www.top500.org|language=en|access-date=2018-07-17}}</ref>

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

* {{As of|2016|10}}, the [[Folding@home]] network has over 100 petaFLOPS of total computing power.<ref>{{cite web|url=https://folding.stanford.edu/home/closing-in-on-100-petaflops/ |title=Closing in on 100 Petaflops|date=May 11, 2016|publisher=Folding@Home|accessdate=July 17, 2016}}</ref><ref>{{cite web|url=http://fah-web.stanford.edu/cgi-bin/main.py?qtype=osstats2 |title=Folding@home team stats pages |publisher=Folding@Home|accessdate=October 14, 2016}}</ref> It was the first computing project of any kind to cross the 1, 2, 3, 4, and 5 native petaFLOPS milestones. 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
 |accessdate  = December 11, 2008
 |url-status     = dead
 |archiveurl  = https://web.archive.org/web/20090131082202/http://www.scei.co.jp/corporate/release/081106de.html
 |archivedate = January 31, 2009
 |df          = mdy
}}</ref>

* {{As of|2018|06}}, the entire [[BOINC]] network averages about 20&nbsp;petaFLOPS.<ref>{{cite web |url=http://boinc.berkeley.edu/ |title=Computering Power |publisher=BOINC |accessdate=June 15, 2018}}</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 |accessdate=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 |accessdate=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|accessdate=June 15, 2018}}</ref>
* {{As of|2018|06}}, [[Great Internet Mersenne Prime Search|GIMPS]], searching for [[Mersenne prime]]s, is sustaining 306&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 |accessdate=June 15, 2018 |postscript=<!--None--> }}</ref>

===Future developments===
{{Further|Exascale computing}}
In 2008, [[James Bamford]]'s book ''[[The Shadow Factory]]'' reported that [[NSA]] told the [[The Pentagon|Pentagon]] it would need an exaflops computer by 2018.<ref>{{cite book |title=The Shadow Factory |author=James Bamford |author-link=James Bamford |publisher= [[Doubleday (publisher)|Doubleday]] |year=2008 |page=[https://archive.org/details/shadowfactory00bamf/page/339 339] |isbn=978-0-385-52132-1|title-link=The Shadow Factory }}</ref>

Given the current speed of progress, [[supercomputer]]s are projected to reach 1&nbsp;exaFLOPS (EFLOPS) in 2018.<ref>{{Cite web | url=http://singularityhub.com/2012/11/01/the-race-to-a-billion-billion-operations-per-second-an-exaflop-by-2018/ | title=The Race to a Billion Billion Operations per Second: An Exaflop by 2018?| date=November 2012}}</ref> [[Cray, Inc.]] announced in December 2009 a plan to build a 1&nbsp;EFLOPS supercomputer before 2020.<ref>{{cite web|url=http://eetimes.com/news/latest/showArticle.jhtml?articleID=222000288 |title=Cray studies exascale computing in Europe |publisher=Eetimes.com |date= |accessdate=February 9, 2012}}</ref> Erik P. DeBenedictis of [[Sandia National Laboratories]] theorizes that a zettaFLOPS (ZFLOPS) computer is required to accomplish full weather modeling of two week time span.<ref>{{cite book |chapter=Reversible logic for supercomputing |title=Proceedings of the 2nd conference on Computing frontiers |last=DeBenedictis |first=Erik P. |authorlink= |year=2005 |publisher= ACM Press|location= New York, NY|isbn=978-1-59593-019-4 |pages=391–402 |chapterurl=http://portal.acm.org/citation.cfm?id=1062325 }}</ref> Such systems might be built around 2030.<ref>{{cite news |date=April 4, 2008 |title=IDF: Intel says Moore's Law holds until 2029 |work=Heise Online |url=http://www.h-online.com/newsticker/news/item/IDF-Intel-says-Moore-s-Law-holds-until-2029-734779.html |url-status=dead |archiveurl=https://web.archive.org/web/20131208075357/http://www.h-online.com/newsticker/news/item/IDF-Intel-says-Moore-s-Law-holds-until-2029-734779.html |archivedate=December 8, 2013 |df=mdy-all}}</ref>

==Cost of computing==
===Hardware costs===
{| class="wikitable"
|-
! Date
! Approximate cost per GFLOPS
! Approximate cost per GFLOPS ({{Inflation-year|US}} US dollars){{Inflation-fn|US}}
!'''Approximate cost per TFLOPS (2017 US dollars)'''
! Platform providing the lowest cost per GFLOPS 
! Comments
|-
| 1961
| $18.7 billion
| ${{Inflation|US|18.7|1961|r=1|fmt=c}} billion
|${{Inflation|US|18.7|1961|r=1|fmt=c}} trillion
| 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 2.4 microseconds.<ref>{{cite web|url=http://computer-history.info/Page4.dir/pages/IBM.7030.Stretch.dir/ |title=The IBM 7030 (STRETCH) |publisher=Norman Hardy |accessdate=February 24, 2017}}</ref>
|-
| 1984
| $18,750,000
| ${{Inflation|US|18750000|1984|r=-4|fmt=c}}
|$44.2 billion
| [[Cray X-MP]]/48
| $15,000,000 / 0.8&nbsp;GFLOPS
|-
| 1997
| $30,000
| ${{Inflation|US|30000|1997|r=-3|fmt=c}}
|$46,000,000
| 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 |accessdate=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=-1|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) -->
|$1,440,000
| [[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-US$1/MFLOPS computing technology. It won the Gordon Bell Prize in 2000.
|-
| {{sort|2000/05|May 2000}}
| $640
| ${{Inflation|US|649|2000|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) -->
|$922,000
| [[Kentucky Linux Athlon Testbed|KLAT2]]
| KLAT2 was the first computing technology which scaled to large applications while staying under US-$1/MFLOPS.<ref>{{Cite web |url=http://aggregate.org/KLAT2/ |title=Kentucky Linux Athlon Testbed 2 (KLAT2) |website=The Aggregate |date= |accessdate=February 9, 2012}}</ref>
|-
| {{sort|2003/08|August 2003}}
| $82
| ${{Inflation|US|82|2003|fmt=c}}
|$109,000
| KASY0
| KASY0 was the first sub-US$100/GFLOPS computing technology.<ref>{{Cite web |url=http://aggregate.org/KASY0/ |title=KASY0 |website=The Aggregate |date=August 22, 2003 |accessdate=February 9, 2012}}</ref>
|-
| {{sort|2007/08|August 2007}}
| $48
| ${{Inflation|US|48|2007|fmt=c}}
|$57,000
| Microwulf
| As of August 2007, this 26.25&nbsp;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 |accessdate=February 9, 2012 |url-status=dead |archiveurl=https://web.archive.org/web/20070912061302/http://www.calvin.edu/~adams/research/microwulf/ |archivedate=September 12, 2007 }}</ref>
|-
| {{sort|2011/03|March 2011}}
| $1.80
| ${{Inflation|US|1.82|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) -->
|$1,980
| 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}}
| $0.75
| ${{Inflation|US|0.75|2012|r=2|fmt=c}}
|$800
| Quad AMD Radeon 7970&nbsp;GHz 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}}
| $0.22
| ${{Inflation|US|0.22|2013|r=2|fmt=c}}
|$230
| Sony PlayStation 4
| The Sony [[PlayStation 4]] is listed as having a peak performance of 1.84&nbsp;TFLOPS, at a price of $400<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}}
| $0.16
| ${{Inflation|US|0.16|2013|r=2|fmt=c}}
|$170
| AMD Sempron 145 & 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 6.771&nbsp;TFLOPS for a total cost of $1090.66.<ref>{{Cite web | url=http://www.freezepage.com/1384601420XCIGYKCBKJ | title=FreezePage}}</ref>
|-
| {{sort|2013/12|December 2013}}
| $0.12
| ${{Inflation|US|0.12|2013|r=2|fmt=c}}
|$130
| Pentium G550 & 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 4.848&nbsp;TFLOPS grand total of US$681.84.<ref>{{Cite web | url=http://www.freezepage.com/1387480124PSLSILVCMJ | title=FreezePage}}</ref>
|-
| {{sort|2015/01|January 2015}}
| $0.08
| ${{Inflation|US|0.08|2015|r=2|fmt=c}}
|$80
| Celeron G1830 & 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 11.5&nbsp;TFLOPS at a grand total of US$902.57.<ref>{{Cite web | url=http://www.freezepage.com/1420850340WGSMHXRBLE | title=FreezePage}}</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=2014-04-08}}</ref>
|-
| {{sort|2017/07|June 2017}}
| $0.06
| ${{Inflation|US|0.06|2017|r=2|fmt=c}}
|$60
| AMD Ryzen 7 1700 & AMD Radeon Vega Frontier Edition
| Built using commercially available parts. AMD Ryzen 7 1700 CPU combined with AMD Radeon Vega FE cards in CrossFire tops out at over 50&nbsp;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
|$0.03
|${{Inflation|US|0.03|2017|r=2|fmt=c}}
|$30
|Intel Celeron G3930 & AMD RX Vega 64
|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 - PCPartPicker|website=pcpartpicker.com|access-date=2017-09-13}}</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 |accessdate=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 |accessdate=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|accessdate=July 8, 2008|archiveurl = https://web.archive.org/web/20080627111128/http://www.ambric.com/pdf/MPR_Ambric_Article_10-06_204101.pdf |archivedate = 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|accessdate=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</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|accessdate=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=24em}}
* [[Computer performance by orders of magnitude]]
* [[Performance per watt#FLOPS per watt]]
* [[Gordon Bell Prize]]
* [[Multiply–accumulate operation]]
* [[LINPACK benchmarks]]
* [[SPECfp]]
* [[SPECint]]
* [[SUPS]]
* [[TOP500]]
* [[Moore's law]]
{{div col end}}
{{clear}}

==References==
{{Reflist|30em|refs=
}}

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

{{Authority control}}

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