Got a YouTube account?

New: enable viewer-created translations and captions on your YouTube channel!

English subtitles

← CPUs Are Not Getting Faster - Intro to Parallel Programming

Get Embed Code
2 Languages

Showing Revision 6 created 05/24/2016 by Udacity Robot.

  1. Let's turn to the technology side a little bit.
  2. Ten years ago if I wanted a faster computer,
  3. then what I would do is wait 6 months and then the clock speed of the CPUs I could buy
  4. would be 30% higher than it was,
  5. and I'd go down to the local store and buy a new faster processor.
  6. And that was largely determined by just a faster clock that I'd go from 1 to 2 to 3 GHz.
  7. That's not happening anymore.
  8. How come?
  9. Yeah, so it's interesting.
  10. A lot of people refer to that scaling in CPU performance as Moore's Law,
  11. which is a little bit of a misnomer.
  12. Moore's Law really predicts a growth rate in the number of devices you can fabricate on a chip.
  13. And Moore's Law is actually alive and well.
  14. We can still, every generation of technology, fabricate more devices
  15. and assuming extra chip, that's still growing exponentially.
  16. What stopped around 2005 was what's called the Nard scaling,
  17. which was scaling the voltages that we operate our chips at as we scaled the dimensions
  18. that we make the transistors at.
  19. And this stopped because of weakened current in the devices.
  20. And without going too technically into it, what that meant is that now when we get a new generation of chip,
  21. let's say we produced a line width by from say unit to 0.7 of whatever that unit was.
  22. A recent jump is from 28 nanometers to 20 nanometers.
  23. When we do that, we now get twice as many devices in the same amount of area.
  24. In the old days we would scale the voltage by 0.7
  25. and since the power goes as CV squared, C would go at 0.7 the voltage will go by 0.7 squared.
  26. We'd wind up with the, I should say energy goes to CV squared.
  27. We would wind up with energy of switching the device being about a third of what it was before √8,
  28. so we'd get this 3x improvement in performace per watt of a basic unit,
  29. and then you can take that and use it in various ways.
  30. One of the ways we used it was to crank clock rate up.
  31. Now that's not happening anymore.
  32. Now that we're holding voltage constant, we're getting a little bit of energy gain from the 0.7 in capacity.
  33. But even that, we're not really getting the whole 0.7, we're getting maybe half of that.
  34. We get perhaps in a generation of technology a 15% improvement in the basic underlying technology.
  35. So that's why you can't take the same old serial processor
  36. and just have it go faster anymore.
  37. You're not going to get any of the frequency improvement.
  38. Frequencies are largely flattened out.
  39. You get more parallelism but it's no longer that factor of 3 each generation that you used to get.
  40. It's now you know perhaps 15% or so before you start applying architecture and circuit innovation.
  41. If you just relied on process, that's all you would get.
  42. Even though each one of those transistors is a little smaller and sucks up a little less power,
  43. that's not nearly enough to change the fact that you're making these bigger
  44. and more parallel every generation.
  45. That's right. The better way to think about it is suppose you got the whole 0.7 on capacitance in that scaling.
  46. That means you now sort of are 30% less energy.
  47. So you can now put 30% more units on.
  48. You would only grow 30% more parallelism in a power constrained environment on that same dime.
  49. You could take that 30% either in clock rate or in more units.
  50. Although it's harder to take in clock rate because that scaling is less linear.
  51. Now, you don't get the whole 30%. You may get half of that for various reasons.
  52. But that's what you're getting to play with each generation until you start innovating.
  53. That's why it's actually kind of fun to be a computer architect these days,
  54. because that's where the value is.
  55. It used be largely in process, and so companies that had a proprietary process kind of an advantage ,
  56. and they still have a little bit of an advantage
  57. but that advantage has really shrunk since process matters less these days.
  58. And what you do with the process, the architecture and circuits
  59. and programming system matters a lot more.