Your SlideShare is downloading. ×
  • Like

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×

Now you can save presentations on your phone or tablet

Available for both IPhone and Android

Text the download link to your phone

Standard text messaging rates apply

Koomey on why ultra-low power computing will change everything

  • 737 views
Published

This talk summarizes the implications of long-term trends in the efficiency of computing, communications, energy storage, and energy harvesting. It's one of my favorites! It took place on October …

This talk summarizes the implications of long-term trends in the efficiency of computing, communications, energy storage, and energy harvesting. It's one of my favorites! It took place on October 31, 2012.

Published in Technology
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
No Downloads

Views

Total Views
737
On SlideShare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
7
Comments
0
Likes
1

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. Copyright  Jonathan  Koomey  2012   SOFTWARE  &  SYSTEMS    DESIGN   Why  ultra-­‐low  power  compu1ng   will  change  everything   Jonathan  Koomey   Research  Fellow,  Steyer-­‐Taylor  Center  for  Energy   Policy  and  Finance,  Stanford  University   jgkoomey@stanford.edu   hSp://www.koomey.com   ARM  Tech  Con,  San  Jose,  CA   October  31,  2012  
  • 2. Copyright  Jonathan  Koomey  2012   RevoluXon   2  
  • 3. Copyright  Jonathan  Koomey  2012   3   Cheap   Smart   Small   Connected   Low  power  +   Self  powered  
  • 4. Copyright  Jonathan  Koomey  2012   CONSIDER  THE  POSSIBILITIES   hSp://proteusdigitalhealth.com/technology/  
  • 5. Copyright  Jonathan  Koomey  2012   Research  quesXon:    How  has  the  energy  efficiency   of  compuXng  changed  over  Xme?   5  
  • 6. Copyright  Jonathan  Koomey  2012   MOORE’S  LAW   •  Not  a  “law”  but  an  empirical  observaXon   about  components/chip     – 1965:    doubling  every  year   – 1975:  doubling  every  2  years   •  Characterizes  economics  of  chip   producXon,  not  physical  limits   •  Ofen  imprecisely  cited,  interpretaXons   changed  over  Xme  (Mollick  2006)  
  • 7. Copyright  Jonathan  Koomey  2012   MOORE’S  ORIGINAL  GRAPH   7  
  • 8. Copyright  Jonathan  Koomey  2012   TRANSISTORS/CHIP  (000S)   The  doubling  Xme  from  1971  to  2006  is  about  1.8  years.    Data  source:  James  Larus,  Microsof  CorporaXon.   8  
  • 9. Copyright  Jonathan  Koomey  2012   How  to  measure  the  energy  efficiency  of   computaXon?   9  
  • 10. Copyright  Jonathan  Koomey  2012   METHOD   •  ComputaXons  per  kWh  =  
  • 11. Copyright  Jonathan  Koomey  2012   DATA   •  Performance  from  Nordhaus  (2007)  or   normalized  to  that  source  using   benchmarks  for  more  recent  computers   •  Used  measured  power  data,  either   published  (e.g.  Weik  1955,  1961,  1964)  or   from  archival  or  recent  computers   – with  computer  fully  uXlized   – with  screen  power  subtracted  for  portables  
  • 12. Copyright  Jonathan  Koomey  2012   Doubling  Xme  for  performance  per   computer  =  1.5  years  in  the  PC  era   12  
  • 13. Copyright  Jonathan  Koomey  2012   PERFORMANCE  TRENDS  (2):   COMPUTATIONS/S/COMPUTER   Source:  Nordhaus  (2007)   with  addiXonal  data   added  by  Koomey  (2009b)  
  • 14. Copyright  Jonathan  Koomey  2012   REAL  POWER  MEASUREMENTS   OF  REAL  COMPUTERS  
  • 15. Copyright  Jonathan  Koomey  2012   COMPUTING  EFFICIENCY   • Doubling  about  every   year  and  a  half  since  the   1940s   • 100x  improvement  every   decade   • Enabled  the  existence  of   laptops  and  smart  phones  
  • 16. Copyright  Jonathan  Koomey  2012   GOOD  CORRELATION,  CLEAR  RESULTS   •  R2  for  computaXons/kWh   – 0.983  for  all  computers,  1946-­‐2009   – 0.970  for  PCs,  1975-­‐2009   •  Doubling  Xme  for  computaXons/kWh   – All  computers:  1.6  years   – PCs:  1.5  years   – Vacuum  tubes:  1.35  years   •  Big  jump  from  tubes  to  transistors  
  • 17. Copyright  Jonathan  Koomey  2012   IMPLICATIONS   •  AcXons  taken  to  improve  performance  also   improve  computaXons  per  kWh   –  Transistors:  Smaller,  shorter  distance  source  to   drain,  fewer  electrons   –  Tubes:    Smaller,  less  capacitance,  lower  currents     •  Trends  make  mobile  and  distributed   compuXng  ever  more  feasible  (baSery  life  up   100x  per  decade  at  constant  compuXng   power)      
  • 18. Copyright  Jonathan  Koomey  2012   LAPTOPS  GROWING  FAST  (WORLD   INSTALLED  BASE,  MILLIONS)   Sources—1985:  Arstechnica  +  Koomey  calcs    1996-­‐2008:  IDC  
  • 19. Copyright  Jonathan  Koomey  2012   IT’S  NOT  JUST  ABOUT  COMPUTING   EFFICIENCY     •  Low  power  >  high  efficiency     •  RevoluXon  is  being  driven  by  the   confluence  of  trends  allowing  low-­‐power   – compuXng   – communicaXons   – sensors   – controls   •  Energy  harvesXng  and  storage  also  criXcal   •  Idle  modes  more  important  than  acXve  
  • 20. Copyright  Jonathan  Koomey  2012   REDUCE  AREA  UNDER  THE  CURVE  
  • 21. Copyright  Jonathan  Koomey  2012   A  MICROCONTROLLER  “RACE”   Source: Eduardo Montanez, Freescale Semiconductor (microcontroller is ARM  ®  Cortex™-­‐m0+  core)
  • 22. Copyright  Jonathan  Koomey  2012   EFFICIENCY  OF  SIMPLE  CELL  PHONES   OVER  TIME  
  • 23. Copyright  Jonathan  Koomey  2012   AN  EXAMPLE  OF  MOBILE  COMPUTING  +   COMMUNICATIONS  ENABLED  BY  EFFICIENCY   http://www.bigbellysolar.com • Compacts trash 5 x • Sends text message when full • PV panel generates power from sunlight • An economic and environmental home run
  • 24. Copyright  Jonathan  Koomey  2012   HOW  ABOUT  REALLY  LOW  POWER   SENSORS?   •  Consider  the  wireless  no-­‐baSery   sensors  created  by  Joshua  R.   Smith  of  the  University  of   Washington   hSp://www.nyXmes.com/2010/07/18/business/18novel.html,   hSp://www.economist.com/node/16295708   –  sensors  use  60  microwaSs  on   average  (60x10-­‐6  waSs)   –  scavenge  power  from  radio  and  TV   signals   •  Other  possible  power  sources   for  similar  devices:    light,  heat,   moXon,  blood  sugar,  digesXve   fluids   Images courtesy of Josh Smith, U of WA
  • 25. Copyright  Jonathan  Koomey  2012   UNIVERSITY  OF  MICHIGAN  MICRO-­‐MOTE   Slide  courtesy  of  David  Blaauw  and  Dennis  Sylvester,  U  of  MI     P=11nW  sleep,  40  μW  acXve,  ARM®  M0  core  
  • 26. Copyright  Jonathan  Koomey  2012   TUMOR  PRESSURE  MONITORING   Slide  courtesy  of  David  Blaauw  and  Dennis  Sylvester,  U  of  MI    
  • 27. Copyright  Jonathan  Koomey  2012   STREETLINE  NETWORKS:  SMART   PARKING,  SMART  CITIES   Slide  courtesy  of  Mark  Noworolski,  Streetline  Networks   Motes  use     <400μW  on     average.    For  LA,   With  40,000  parking   spots,  that  implies   total  mote  power   of  about  15W.   Mote  technology   is  from  Dust     Networks  
  • 28. Copyright  Jonathan  Koomey  2012   STREETLINE  NETWORKS:  SMART   PARKING,  SMART  CITIES   Slide  courtesy  of  Mark  Noworolski,  Streetline  Networks  
  • 29. Copyright  Jonathan  Koomey  2012   STREETLINE  NETWORKS:     VARIABLE  PARKING  SIGNS   Slide  courtesy  of  Mark  Noworolski,  Streetline  Networks  
  • 30. Copyright  Jonathan  Koomey  2012   DEEPER  IMPLICATIONS   • Move  bits,  not  atoms   • Customized  data  collecXon  (focus  on  nanodata,   not  big  data)   • Ever  more  precise  control  of  processes   • Real-­‐Xme  analysis   • Enabling  “the  internet  of  things”   • Bo#om  line:    beSer  matching  of  energy  services   demanded  with  those  supplied,  beSer  real-­‐Xme   control,  and  beSer  analysis.  
  • 31. Copyright  Jonathan  Koomey  2012   THESE  TRENDS  STILL  HAVE  A  LONG   WAY  TO  RUN   Psssst:    Researchers  at   Purdue  and  the  University   of  New  South  Wales   recently  created  a  reliable   one  atom  transistor…   2041  
  • 32. Copyright  Jonathan  Koomey  2012   BIG  UNANSWERED  QUESTIONS   •  Could  we  do  beSer  than  historical  trends?   •  Might  we  do  worse?    If  so,  why?   •  What’s  next  as  we  approach  theoreXcal   limits?  
  • 33. Copyright  Jonathan  Koomey  2012   CONCLUSIONS   •  QuanXtaXve  results   –  In  the  PC  era  (1976-­‐2009)  performance  per  computer   and  computaXons  per  kWh  doubled  every  1.5  years   –  From  ENIAC  to  the  present,  computaXons  per  kWh   doubled  every  1.6  years   •  Performance  and  efficiency  improvements   inextricably  linked   •  SXll  far  from  theoreXcal  limits   •  Big  implicaXons  for  mobile  technologies   •  The  future  belongs  to  low  power  systems!  
  • 34. Copyright  Jonathan  Koomey  2012   Viva  la  Revolución!   34  
  • 35. Copyright  Jonathan  Koomey  2012   THANKS!   •  Rob  Bernard,  Microsof  (funder)   •  Lorie  Wigle,  Intel  (funder)   •  Stephen  Berard  of  Microsof  (coauthor)   •  Marla  Sanchez  of  LBNL  and  CMU  (coauthor),  and   •  Henry  Wong  of  Intel  (coauthor)   •  The  Computer  History  Museum   •  IEEE  Annals  of  the  History  of  Compu8ng  
  • 36. Copyright  Jonathan  Koomey  2012   FURTHER  READING   Koomey,  Jonathan  G.,  Stephen  Berard,  Marla  Sanchez,  and  Henry  Wong.  2011.   "ImplicaXons  of  Historical  Trends  in  the  Electrical  Efficiency  of  CompuXng."     IEEE  Annals  of  the  History  of  Compu8ng.    vol.  33,  no.  3.  July-­‐September.  pp.   46-­‐54.  [hSp://doi.ieeecomputersociety.org/10.1109/MAHC.2010.28]   Koomey,    Jonathan.  2012.  "The  CompuXng  Trend  that  Will  Change  Everything."  In   Technology   Review.   April   2.   [hSp://www.technologyreview.com/news/ 427444/the-­‐compuXng-­‐trend-­‐that-­‐will-­‐change-­‐everything/]   Greene,  Kate.  2011.  "A  New  and  Improved  Moore's  Law."  In  Technology  Review.   September   12.   [hSp://www.technologyreview.com/compuXng/38548/? p1=A1]   Eisenberg,  Anne.  2010.  "Bye-­‐Bye  BaSeries:  Radio  Waves  as  a  Low-­‐Power  Source."   The   New   York   Times.     New   York,   NY.     July   18.   p.   BU3.   [hSp:// www.nyXmes.com/2010/07/18/business/18novel.html]  
  • 37. Copyright  Jonathan  Koomey  2012   REFERENCES   •  Feynman,  Richard  P.  2001.  The  Pleasure  of  Finding  Things  Out:    The  Best  Short  Works  of  Richard  P.   Feynman.  London,  UK:  Penguin  Books.     •  Hilbert,  MarXn,  and  Priscila  López.  2011.  "The  World's  Technological  Capacity  to  Store,   Communicate,  and  Compute  InformaXon."    Science.    vol.  332,  no.  6025.  April  1.  pp.  60-­‐65.     •  Koomey,  Jonathan.  2008.  "Worldwide  electricity  used  in  data  centers."    Environmental  Research   Le#ers.  vol.  3,  no.  034008.  September  23.  <h#p://stacks.iop.org/1748-­‐9326/3/034008>.   •  Koomey,  Jonathan  G.,  ChrisXan  Belady,  Michael  PaSerson,  Anthony  Santos,  and  Klaus-­‐Dieter  Lange.   2009a.  Assessing  trends  over  8me  in  performance,  costs,  and  energy  use  for  servers.  Oakland,  CA:   AnalyXcs  Press.    August  17.  <hSp://www.intel.com/pressroom/kits/ecotech>.       •  Koomey,  Jonathan  G.,  Stephen  Berard,  Marla  Sanchez,  and  Henry  Wong.  2011.  "ImplicaXons  of   Historical  Trends  in  The  Electrical  Efficiency  of  CompuXng."    IEEE  Annals  of  the  History  of   Compu8ng.    vol.  33,  no.  3.  July-­‐September.  pp.  2-­‐10.    <hSps://files.me.com/jgkoomey/u0zi7l>   •  Koomey,  Jonathan.  2011.  Growth  in  data  center  electricity  use  2005  to  2010.  Oakland,  CA:  AnalyXcs   Press.    August  1.  <hSp://www.analyXcspress.com/datacenters.html>   •  Mollick,  Ethan.  2006.  "Establishing  Moore’s  Law."    IEEE  Annals  of  the  History  of  Compu8ng   (Published  by  the  IEEE  Computer  Society).    July-­‐September.  pp.  62-­‐75.    
  • 38. Copyright  Jonathan  Koomey  2012   REFERENCES  (2)   •  Yoonmyung,  Lee,  Kim  Gyouho,  Bang  Suyoung,  Kim  Yejoong,  Lee  Inhee,  P.  DuSa,  D.  Sylvester,  and  D.   Blaauw.  2012.  A  modular  1mm3  die-­‐stacked  sensing  pla]orm  with  op8cal  communica8on  and   mul8-­‐modal  energy  harves8ng.  Proceedings  of  the  Solid-­‐State  Circuits  Conference  Digest  of   Technical  Papers  (ISSCC),  2012  IEEE  InternaXonal.    19-­‐23  Feb.  2012.    [hSp://ieeexplore.ieee.org/xpl/ mostRecentIssue.jsp?punumber=6171933]     •  Moore,  Gordon  E.  1965.  "Cramming  more  components  onto  integrated  circuits."  In  Electronics.  April   19.     •  Moore,  Gordon  E.  1975.  "Progress  in  Digital  Integrated  Electronics."    IEEE,  IEDM  Tech  Digest.    pp.   11-­‐13.  <hSp://www.ieee.org/>   •  Nordhaus,  William  D.  2007.  "Two  Centuries  of  ProducXvity  Growth  in  CompuXng."    The  Journal  of   Economic  History.    vol.  67,  no.  1.  March.  pp.  128-­‐159.  <hSp://nordhaus.econ.yale.edu/ recent_stuff.html>   •  Weik,  MarXn  H.  1955.  A  Survey  of  Domes8c  Electronic  Digital  Compu8ng  Systems.  Aberdeen   Proving  Ground,  Maryland:  BallisXc  Research  Laboratories.  Report  No.  971.    December.  <hSp://ed-­‐ thelen.org/comp-­‐hist/BRL.html>   •  Weik,  MarXn  H.  1961.  A  Third  Survey  of  Domes8c  Electronic  Digital  Compu8ng  Systems.  Aberdeen   Proving  Ground,  Maryland:  BallisXc  Research  Laboratories.  Report  No.  1115.    March.  <hSp://ed-­‐ thelen.org/comp-­‐hist/BRL61.html>   •  Weik,  MarXn  H.  1964.  A  Fourth  Survey  of  Domes8c  Electronic  Digital  Compu8ng  Systems   (Supplement  to  the  Third  Survey).  Aberdeen  Proving  Ground,  Maryland:  BallisXc  Research   Laboratories.  Report  No.  1227.    January.  <hSp://ed-­‐thelen.org/comp-­‐hist/BRL64.html>