When will supercomputers take over the world?

A talk I gave at Nerd Nite in Oakland (eastbay.nerdnite.com). October, 2013.

Transcript

1. When  will  supercomputers  take  over  the  world?  (Never.)   Paul  G.  Constantine,  @DrPaulynomial   Assistant  Professor,  Colorado  School  of  Mines   Ph.D.  Computational  and  Mathematical  Engineering,  Stanford  (2009)  
2. A  supercomputer  is  a  bunch  of  regular     computers  connected  together.   Lawrence  Livermore  National  Lab’s  Sequoia   1,572,864  processor  cores  
3. Supercomputers  are  kinda  ‘spensive.   •  $100-­‐to-­‐250  million  for  design  and  assembly   •  $6-­‐to-­‐7  million  per  year  for  power   http://techland.time.com/2012/06/19/what-­‐exactly-­‐is-­‐a-­‐supercomputer/  
4. Performance  is  measured  by  solving     a  big  system  of  linear  equations.   3 x + 2 y = 7 x 4 y = 2 Two  equations  and  two  variables   flops  =  floating  point  operations  per  second   MEGA   GIGA   TERA   PETA   EXA   106   109   1012   1015   1018  
5. www.top500.org  
6. www.top500.org  
7. Getting  to  EXAFLOPS  is  going  to  be  difficult.   •  Processors  have  stopped  getting  faster.   •  There  is  an  ambitious  power  target  (20MW).   •  With  all  those  computers,  failures  will  be  frequent.  
8. What  do  the  supercomputers  do?  
9. Computational  science  is  more     than  just  computer  science.   •  Computational  science  is  doing  science  with  a  computer.   •  Computer  science  is  the  science  of  the  computer.  
10. Computers  replace  and/or     complement  experiments.   If  experiments  are  too   expensive  or  illegal…   If  you  need  to  run  lots  and   lots  of  experiments…  
11. Several  scientific  fields  use  supercomputers.   Astrophysics   Climate  &  weather   Turbulence  and  acoustics   Oil  reservoirs   Biological   systems  
13. “REALITY”   MATH     MODEL   (think:  differential  equations)   (think:  measurements)  
14. “REALITY”   MATH     MODEL   (think:  differential  equations)   SIMPLER   MATH     MODEL   Approximation   (think:  measurements)  
15. “REALITY”   MATH     MODEL   (think:  differential  equations)   COMPUTER     MODEL   Approximation   (think:  measurements)  
16. Powerful  computers  approximate  better.   Truth Approximation $  
17. Powerful  computers  approximate  better.   Truth Approximation $$  
18. Powerful  computers  approximate  better.   Truth Approximation $$$  
19. Powerful  computers  approximate  better.   Truth Approximation $$$$  
20. Powerful  computers  approximate  better.   Truth Approximation $$$$$  
21. Powerful  computers  approximate  better.   Truth Approximation $$$$$$  
22. Impress  your  friends  by  using  the   word  multiscale  in  a  sentence.   We  need  the  supercomputers  to   resolve  the  multiscale  phenomena   found  in  most  physical  systems.  
23. Questions?   What  does  it  mean  for  an   algorithm  to  be  “parallel?”     What  if  the  original  math  model   isn’t  any  good?     Does  1,000  petascale  computers   count  as  an  exascale  computer?     Can  a  supercomputer  predict   [insert  thing  to  predict]?     Can  I  take  your  class?   Paul  G.  Constantine,  @DrPaulynomial,  http://inside.mines.edu/~pconstan