View source for Performance per watt

{{Short description|Computer energy efficiency}}
{{redirect-distinguish|Power efficiency|Mechanical efficiency}}
{{use dmy dates |date=February 2021}}

In [[computing]], '''performance per watt''' is a measure of the [[Energy conversion efficiency|energy efficiency]] of a particular [[computer architecture]] or [[computer hardware]]. Literally, it measures the rate of computation that can be delivered by a computer for every [[watt]] of power consumed. This rate is typically measured by performance on the [[LINPACK]] benchmark when trying to compare between computing systems: an example using this is the [[Green500]] list of supercomputers. Performance per watt has been suggested to be a more sustainable measure of computing than [[Moore's law|Moore's Law]].<ref>{{Cite web|last1=Aitken|first1=Rob|last2=Fellow|last3=Technology|first3=Director of|last4=Arm|date=2021-07-12|title=Performance per Watt Is the New Moore's Law|url=https://www.arm.com/blogs/blueprint/performance-per-watt|access-date=2021-07-16|website=Arm Blueprint|language=en-US}}</ref>

System designers building [[parallel computing|parallel computers]], such as [[Google search technology#Production hardware|Google's hardware]], pick CPUs based on their performance per watt of power, because the cost of powering the CPU outweighs the cost of the CPU itself.<ref>[http://news.cnet.com/Power+could+cost+more+than+servers,+Google+warns/2100-1010_3-5988090.html Power could cost more than servers, Google warns], CNET, 2006</ref>

Spaceflight computers have hard limits on the maximum power available and also have hard requirements on minimum real-time performance. A ratio of processing speed to required electrical power is more useful than raw processing speed.<ref name="sc-7">
D. J. Shirley; and M. K. McLelland.
[https://digitalcommons.usu.edu/cgi/viewcontent.cgi?article=2656&context=smallsat "The Next-Generation SC-7 RISC Spaceflight Computer"].
p. 1, 2.
</ref>

==Definition==
The performance and power consumption metrics used depend on the definition; reasonable measures of performance are [[FLOPS]], [[Instructions per second|MIPS]], or the score for any [[Benchmark (computing)|performance benchmark]].  Several measures of power usage may be employed, depending on the purposes of the metric; for example, a metric might only consider the electrical power delivered to a machine directly, while another might include all power necessary to run a computer, such as cooling and monitoring systems. The power measurement is often the average power used while running the benchmark, but other measures of power usage may be employed (e.g. peak power, idle power).

For example, the early [[UNIVAC I]] computer performed approximately 0.015 operations per watt-second (performing 1,905 operations per second (OPS), while consuming 125&nbsp;kW). The [[Fujitsu]] [[FR-V]] [[VLIW]]/[[vector processor]] [[system on a chip]] in the 4 FR550 core variant released 2005 performs 51 Giga-OPS with 3 watts of power consumption resulting in 17 billion operations per watt-second.<ref>{{cite press release |author=<!--Staff writer(s); no by-line.--> |date=2020-02-07 |title=Fujitsu Develops Multi-core Processor for High-Performance Digital Consumer Products |url=https://www.fujitsu.com/global/about/resources/news/press-releases/2005/0207-01.html |url-status=live |publisher=Fujitsu |archive-url=https://web.archive.org/web/20190325233011/http://www.fujitsu.com/global/about/resources/news/press-releases/2005/0207-01.html |archive-date=2019-03-25 |access-date=2020-08-08}}</ref><ref>[http://www.fujitsu.com/downloads/MAG/vol42-2/paper03.pdf FR-V Single-Chip Multicore Processor:FR1000] {{webarchive|url=https://web.archive.org/web/20150402150434/http://www.fujitsu.com/downloads/MAG/vol42-2/paper03.pdf |date=2015-04-02 }} Fujitsu</ref> This is an improvement by over a trillion times in 54 years.

Most of the power a computer uses is converted into heat, so a system that takes fewer watts to do a job will require less cooling to maintain a given [[operating temperature]].  Reduced cooling demands makes it easier to [[quiet PC|quiet a computer]]. Lower energy consumption can also make it less costly to run, and reduce the environmental impact of powering the computer (see [[green computing]]).
If installed where there is limited [[climate control]], a lower power computer will operate at a lower temperature, which may make it more reliable.  In a climate controlled environment, reductions in direct power use may also create savings in climate control energy.

Computing energy consumption is sometimes also measured by reporting the energy required to run a particular benchmark, for instance [[EEMBC]] EnergyBench.  Energy consumption figures for a standard workload may make it easier to judge the effect of an improvement in [[Electrical efficiency|energy efficiency]].

When performance is defined as {{Sfrac|operations|[[second]]}}, then performance per watt can be written as {{Sfrac|operations|[[watt-second]]}}. Since a watt is one {{Sfrac|[[joule]]|second}}, then performance per watt can also be written as {{Sfrac|operations|joule}}.

==FLOPS per watt==
{{anchor|MFLOPS PER WATT|GFLOPS PER WATT}}
[[File:Green500 evolution.svg|thumb|400px|Exponential growth of supercomputer performance per watt based on data from the [[Green500]] list. The red crosses denote the most power efficient computer, while the blue ones denote the computer ranked#500.]]
'''FLOPS per watt''' is a common measure. Like the [[FLOPS]] ([[floating point|Floating Point]] Operations Per Second) metric it is based on, the metric is usually applied to [[scientific computing]] and simulations involving many [[floating point]] calculations.

===Examples===
{{As of|2016|06}}, the Green500 list rates the two most efficient supercomputers highest{{snd}} those are both based on the same [[Manycore processor|manycore]] accelerator [[PEZY-SCnp]] Japanese technology in addition to Intel Xeon processors<!--computers both named ZettaScaler-1.6-->{{snd}} both at [[RIKEN]], the top one at 6673.8&nbsp;MFLOPS/watt; and the third ranked is the Chinese-technology [[Sunway TaihuLight]] (a much bigger machine, that is the ranked 2nd on [[TOP500]], the others are not on that list) at 6051.3&nbsp;MFLOPS/watt.<ref>{{cite web |url=https://www.top500.org/green500/lists/2016/06/ |title=Green500 List for June 2016}}</ref>

In June 2012, the Green500 list rated [[Blue Gene/Q|BlueGene/Q, Power BQC 16C]] as the most efficient supercomputer on the TOP500 in terms of FLOPS per watt, running at 2,100.88&nbsp;MFLOPS/watt.<ref>{{cite web | url = http://www.green500.org/lists/2012/06/top/list.php | work = Green500 | title = The Green500 List | url-status = dead | archive-url = https://web.archive.org/web/20120703053827/http://www.green500.org/lists/2012/06/top/list.php | archive-date = 2012-07-03 }}</ref>

In November 2010, IBM machine, [[IBM Blue Gene#Blue Gene/Q|Blue Gene/Q]] achieves 1,684&nbsp;MFLOPS/watt.<ref>{{cite web| url=http://www.serverwatch.com/hreviews/article.php/3913536/Top500-Supercomputing-List-Reveals-Computing-Trends.htm| title = Top500 Supercomputing List Reveals Computing Trends| date = 20 July 2010|quote=IBM... BlueGene/Q system .. setting a record in power efficiency with a value of 1,680&nbsp;Mflops/watt, more than twice that of the next best system.}}</ref><ref>{{cite web| url=http://www.datacenterknowledge.com/archives/2010/11/18/ibm-system-clear-winner-in-green-500/|title = IBM Research A Clear Winner in Green 500|date = 2010-11-18}}</ref>

On 9 June 2008, CNN reported that [[IBM Roadrunner|IBM's Roadrunner]] supercomputer achieves 376&nbsp;MFLOPS/watt.<ref>{{cite news | url = http://www.cnn.com/2008/TECH/06/09/fastest.computer.ap/index.html  | work = CNN | title = Government unveils world's fastest computer |quote= performing 376 million calculations for every watt of electricity used. |archive-url = https://web.archive.org/web/20080610155646/http://www.cnn.com/2008/TECH/06/09/fastest.computer.ap/index.html |archive-date = 2008-06-10}}</ref><ref>{{cite web|url = http://www.hpcwire.com/topic/processors/IBM_Roadrunner_Takes_the_Gold_in_the_Petaflop_Race.html|title = IBM Roadrunner Takes the Gold in the Petaflop Race|url-status = dead|archive-url = https://web.archive.org/web/20080613131535/http://www.hpcwire.com/topic/processors/IBM_Roadrunner_Takes_the_Gold_in_the_Petaflop_Race.html|archive-date = 2008-06-13}}</ref>

As part of the [[Intel Tera-Scale]] research project, the team produced an 80-core CPU that can achieve over 16,000&nbsp;MFLOPS/watt.<ref>{{cite web | url = http://www.tgdaily.com/content/view/30929/135/ | work = TG Daily | title = Intel squeezes 1.8 TFlops out of one processor | url-status = dead | archive-url = https://web.archive.org/web/20071203023419/http://www.tgdaily.com/content/view/30929/135/ | archive-date = 2007-12-03 }}</ref><ref>{{cite web | url = http://techresearch.intel.com/articles/Tera-Scale/1449.htm | work = Intel Technology and Research | title = Teraflops Research Chip}}</ref> The future of that CPU is not certain.

Microwulf, a low cost desktop [[Beowulf (computing)|Beowulf cluster]] of four dual-core [[Athlon 64 X2]] 3800+ computers, runs at 58&nbsp;MFLOPS/watt.<ref name="microwulf">{{cite web | url = http://www.calvin.edu/~adams/research/microwulf/power/ | title = Microwulf: Power Efficiency | author = Joel Adams | work = Microwulf: A Personal, Portable Beowulf Cluster }}</ref>

Kalray has developed a 256-core VLIW CPU that achieves 25,000&nbsp;MFLOPS/watt. Next generation is expected to achieve 75,000&nbsp;MFLOPS/watt.<ref>{{cite web | url = http://www.kalray.eu/products/mppa-manycore/ | title = MPPA MANYCORE - Many-core processors - KALRAY - Agile Performance}}</ref> However, in 2019 their latest chip for embedded is 80-core and claims up to 4&nbsp;TFLOPS at 20&nbsp;W.<ref>{{Cite web|url=https://www.kalray.eu/kalray-announces-the-tape-out-of-coolidge-on-tsmc-16nm-process-technology/|title=Kalray announces the Tape-Out of Coolidge on TSMC 16NM process technology|date=2019-07-31|website=Kalray|language=en-US|access-date=2019-08-12}}</ref>

[[Adapteva]] announced the [[Adapteva#Epiphany V|Epiphany V]], a 1024-core 64-bit RISC processor intended to achieve 75&nbsp;GFLOPS/watt,<ref>{{cite web|last1=Olofsson|first1=Andreas|title=Epiphany-V: A 1024-core 64-bit RISC processor|url=https://www.parallella.org/2016/10/05/epiphany-v-a-1024-core-64-bit-risc-processor/|access-date=6 October 2016}}</ref><ref>{{cite web|last1=Olofsson|first1=Andreas|title=Epiphany-V: A 1024 processor 64-bit RISC System-On-Chip|url=https://www.parallella.org/wp-content/uploads/2016/10/e5_1024core_soc.pdf|access-date=6 October 2016}}</ref> while they later announced that the Epiphany&nbsp;V was "unlikely" to become available as a commercial product

US Patent [http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearch-bool.html&r=1&f=G&l=50&co1=AND&d=PTXT&s1=matteo&s2=gravina&OS=matteo+AND+gravina&RS=matteo+AND+gravina 10,020,436], July 2018 claims three intervals of 100, 300, and 600 GFLOPS/watt.

==GPU efficiency==
[[Graphics processing unit]]s (GPU) have continued to increase in energy usage, while CPUs designers have recently{{When|date=August 2024}} focused on improving performance per watt.  High performance GPUs may draw large amount of power, therefore intelligent techniques are required to manage GPU power consumption. Measures like [[3DMark|3DMark2006 score]] per watt can help identify more efficient GPUs.<ref>{{cite web | url=http://www.codinghorror.com/blog/archives/000662.html | title=Video Card Power Consumption | first=Jeff | last=Atwood | date=2006-08-18 | access-date=26 March 2008 | archive-date=8 September 2008 | archive-url=https://web.archive.org/web/20080908060043/http://www.codinghorror.com/blog/archives/000662.html | url-status=dead }}</ref> However that may not adequately incorporate efficiency in typical use, where much time is spent doing less demanding tasks.<ref>{{cite web | url = http://www.xbitlabs.com/articles/video/display/power-noise.html | title = Video card power consumption | work = Xbit Labs | url-status = dead | archive-url = https://web.archive.org/web/20110904054636/http://www.xbitlabs.com/articles/video/display/power-noise.html | archive-date = 2011-09-04 }}</ref>

With modern GPUs, energy usage is an important constraint on the maximum computational capabilities that can be achieved. GPU designs are usually highly scalable, allowing the manufacturer to put multiple chips on the same video card, or to use multiple video cards that work in parallel. Peak performance of any system is essentially limited by the amount of power it can draw and the amount of heat it can dissipate. Consequently, performance per watt of a GPU design translates directly into peak performance of a system that uses that design.

Since GPUs may also be used for some [[GPGPU|general purpose computation]], sometimes their performance is measured in terms also applied to CPUs, such as FLOPS per watt.

==Challenges==
{{Missing information|section|inflationary effect of low clock and power limits, e.g.;<ref>{{Cite web|url=https://www.reddit.com/r/hardware/comments/k3iobs/psa_performance_doesnt_scale_linearly_with/|title=PSA: Performance Doesn't Scale Linearly with Wattage (Aka testing M1 versus a Zen 3 5600X at the same Power Draw)|date=29 November 2020}}</ref> also Energy/Frequency Convexity Rule|date=November 2020}}
While performance per watt is useful, absolute power requirements are also important.  Claims of improved performance per watt may be used to mask increasing power demands.  For instance, though newer generation GPU architectures may provide better performance per watt, continued performance increases can negate the gains in efficiency, and the GPUs continue to consume large amounts of power.<ref>{{cite web | url = http://www.bit-tech.net/columns/2007/05/20/performance_per_what/1 | title = Performance per What? | author = Tim Smalley | work = Bit Tech | access-date = 2008-04-21 }}</ref>

Benchmarks that measure power under heavy load may not adequately reflect typical efficiency.  For instance, 3DMark stresses the 3D performance of a GPU, but many computers spend most of their time doing less intense display tasks (idle, 2D tasks, displaying video).  So the 2D or idle efficiency of the graphics system may be at least as significant for overall energy efficiency.  Likewise, systems that spend much of their time in standby or [[standby power|soft off]] are not adequately characterized by just efficiency under load.  To help address this some benchmarks, like [[SPECpower]], include measurements at a series of load levels.<ref>{{cite web | url = http://blogs.zdnet.com/Ou/?p=927 | archive-url = https://web.archive.org/web/20071216112117/http://blogs.zdnet.com/Ou/?p=927 | url-status = dead | archive-date = 16 December 2007 | work = ZDNet | title = SPEC launches standardized energy efficiency benchmark}}</ref>

The efficiency of some electrical components, such as [[voltage regulator]]s, decreases with increasing temperature, so the power used may increase with temperature.  Power supplies, motherboards, and some video cards are some of the subsystems affected by this.  So their power draw may depend on temperature, and the temperature or temperature dependence should be noted when measuring.<ref>{{cite web | url = http://www.silentpcreview.com/article821-page5.html | title = Asus EN9600GT Silent Edition Graphics Card | author = Mike Chin | page = 5 | work = Silent PC Review | access-date = 2008-04-21}}</ref><ref name="SPCRNewLevels">{{cite web | url = http://www.silentpcreview.com/article814-page1.html | title = 80 Plus expands podium for Bronze, Silver & Gold | author = Mike Chin | work = Silent PC Review | date = 19 March 2008 | access-date = 2008-04-21 }}</ref>

Performance per watt also typically does not include full [[Life cycle assessment|life-cycle costs]].  Since computer manufacturing is energy intensive, and computers often have a relatively short lifespan, energy and materials involved in production, distribution, [[electronic waste|disposal]] and [[computer recycling|recycling]] often make up significant portions of their cost, energy use, and environmental impact.<ref>{{cite web | url=http://www.ecopcreview.com/LCA_and_ECPR | title=Life Cycle Analysis and Eco PC Review | work=Eco PC Review | author=Mike Chin | url-status=dead | archive-url=https://web.archive.org/web/20080304062508/http://www.ecopcreview.com/LCA_and_ECPR | archive-date=2008-03-04 }}</ref><ref>{{cite journal | journal = Environ. Sci. Technol.
| title = Energy intensity of computer manufacturing: hybrid assessment combining process and economic input-output methods | author = Eric Williams | year=2004
| doi = 10.1021/es035152j
| volume = 38
| pages = 6166–74 | pmid = 15573621 | issue = 22| bibcode = 2004EnST...38.6166W}}</ref>

Energy required for climate control of the computer's surroundings is often not counted in the wattage calculation, but it can be significant.<ref name="impLowPower">{{cite journal | journal = CT Watch Quarterly
| title = The Importance of Being Low Power in High Performance Computing
| author = Wu-chun Feng
| url = http://www.ctwatch.org/quarterly/articles/2005/08/the-importance-of-being-low-power-in-high-performance-computing/
| volume = 1 | issue = 5 | year = 2005
}}</ref>

==Other energy efficiency measures==
SWaP (space, wattage and performance) is a [[Sun Microsystems]] metric for [[data center]]s, incorporating power and space:

:<math>\mathrm{SWaP} = \frac{\mathrm{Performance}}{\mathrm{Space} \cdot \mathrm{Power}}</math>

Where performance is measured by any appropriate benchmark, and space is size of the computer.<ref>{{cite web|last=Greenhill|first=David|title=SWaP Space Watts and Power|url=http://www.energystar.gov/ia/products/downloads/Greenhill_Pres.pdf|work=US EPA Energystar|access-date=14 November 2013}}</ref>

Reduction of power, mass, and volume is also important for spaceflight computers.<ref name="sc-7" />

==See also==
; Energy efficiency benchmarks
* [[Average CPU power]] (ACP){{snd}} a measure of power consumption when running several standard benchmarks
* [[EEMBC]]{{snd}} EnergyBench
* [[SPECpower]]{{snd}} a benchmark for web servers running Java (Server Side Java Operations per Joule)

; Other
* [[Data center infrastructure efficiency]] (DCIE)
* [[Energy proportional computing]]
* [[IT energy management]]
* [[Koomey's law]]
* [[Landauer's principle]]
* [[Low-power electronics]]
* [[Power usage effectiveness]] (PUE)
* [[Processor power dissipation]]

==Notes and references==
{{Reflist|30em}}

==Further reading==
* {{cite journal | title = Making a case for Efficient Supercomputing
| author = Wu-Chun Feng
| journal = ACM Queue
| volume = 1 | date = October 2003
| url = http://www.acmqueue.com/modules.php?name=Content&pa=showpage&pid=80&page=1 | issue = 7
}}
* {{cite journal | title = HPC Power Efficiency and the Green500
| author = Kirk W. Cameron
| journal = HPC Wire
| volume = 27 | date = November 2013
| url = http://www.hpcwire.com/2013/11/20/hpc-power-efficiency-green500/ | issue = 11
}}
* {{cite journal | title = The Green500 List: Encouraging Sustainable Supercomputing
| author = Wu-Chun Feng and Kirk W. Cameron
| journal = IEEE Computer
| volume = 40 | date = December 2007
| url = http://www.hpcwire.com/2013/11/20/hpc-power-efficiency-green500/ | issue = 12
}}

==External links==
* [https://www.top500.org/lists/green500/ The Green500]

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

[[Category:Benchmarks (computing)]]
[[Category:Computers and the environment]]
[[Category:Electric power]]
[[Category:Energy conservation]]
[[Category:Computer performance]]