Return to Video

CPU Speed Remaining Flat - Intro to Parallel Programming

  • 0:00 - 0:05
    To understand why the GPU is such an interesting processor today, we'll start with technology trends.
  • 0:05 - 0:07
    Why the world has gone parallel.
  • 0:07 - 0:08
    And first, we'll start with some good news.
  • 0:08 - 0:11
    Modern processors are made from transistors.
  • 0:11 - 0:14
    Each year those transistors get smaller and smaller.
  • 0:14 - 0:18
    This graph is from Stanford's CPUDB project. Thanks, guys.
  • 0:18 - 0:22
    What it shows is the feature size of processors over time,
  • 0:22 - 0:25
    where the feature size is the minimum size of a transistor or wire on a chip.
  • 0:25 - 0:29
    So what we're seeing is this is time going in this direction.
  • 0:29 - 0:32
    That's a long time ago and that's today.
  • 0:32 - 0:35
    This is the feature size, so how big transistors are.
  • 0:35 - 0:38
    Notice that it's getting smaller and smaller every generation.
  • 0:38 - 0:42
    When you hear talk about 45 nanometer process or a 22 nanometer
  • 0:42 - 0:45
    that's what we're referring to, the feature size.
  • 0:45 - 0:48
    We see that it's consistently going down over time.
  • 0:48 - 0:52
    As the feature size decreases, transistors get smaller, run faster, use less power,
  • 0:52 - 0:54
    and we can put more of them on a chip.
  • 0:54 - 0:59
    The consequence is that we have more and more resources for computation every single year.
  • 0:59 - 1:01
    However, we've also got some bad news.
  • 1:01 - 1:04
    Historically, as transistors improved, processor designers would then
  • 1:04 - 1:08
    increase the clock rates of processors, running them faster and faster every year.
  • 1:08 - 1:12
    Let's take a look at this diagram of clock speeds over the years.
  • 1:12 - 1:17
    So again, we have time going on this axis, okay? So, a long time ago, today.
  • 1:17 - 1:22
    Here we have clock frequency—how fast we're clocking these transistors.
  • 1:22 - 1:26
    Historically, one of the primary drivers of clock performance has been clock speed increases.
  • 1:26 - 1:30
    We see over many years, we see clock speeds continue to go up.
  • 1:30 - 1:36
    However, over the last decade we see that clock speeds have essentially remained constant.
タイトル:
CPU Speed Remaining Flat - Intro to Parallel Programming
概説:

more » « less
Video Language:
English
Team:
Udacity
プロジェクト:
CS344 - Intro to Parallel Programming
Duration:
01:37

English subtitles

改訂 Compare revisions