The end of Chapter 1

• If you have not yet done so, READ CHAPTER 1.
• 1.7 the Power Wall
  • They tell us energy is directly related to a load on each transistor (called Capacitive load) and the square of the Voltage
  • Energy ∝ 1/2 Capacitive load × Voltage²
  • Further they tell us that power is
  • Power ∝ 1/2 Capacitive load × Voltage² × Frequency Switched.
  • In this case, Frequency Switched is the clock rate.
  • Typically in each generation, we were able to drop by voltage.
  • Hence the table on page 40 showing an increase in speed but a smaller increase in power.
  • 
  • They say that we have seem to have hit a limit for reducing voltage.
  • So we can't continue to drive our chips faster.
• 1.8, The Sea of Change.
  • So what have we been doing with the extra transistors?
  • Multiple Cores
  • Is this a good solution?
    • As I type on my machine I am
      • using an editor (very few cycles really)
      • listening to Pandora (a few more cycles)
      • running my browser.
      • And my xserver, and other things...
      • running a web server (early Sunday afternoon, no one is using it)
    • All of these tasks can run on a different core
    • 
    • And the easiest type of parallel processing is demonstrated here.
  • But what about "faster programs"?
    • Parallel programming is generally much harder than sequential programming.
    • Introduce Flynn's
    • The easiest type is SIMD
      • Single Instruction Multiple Data
      • This is what happens in GPUs.
      • A task (such as drawing a bunch of pixels) is split among a bunch of processors.
      • Only works in very parallelizable situations.
    • The hard type is MIMD in a single task
      • Multiple Instructions Multiple data.
      • Coordinating tasks, data are major problems.
      • Many issues (deadlock, race conditions, ...)
      • Take OS to learn about these.
• 1.9, Benchmarking the Intel Core i7
  • Over the years there has been an attempt to generate a "true" measurement of performance.
  • This would allow consumers to compare processors.
  • SPEC (System Performance Evaluation Cooperative) (or Standard Performance Evaluation Corp)
    • Establishes standards for processor comparison.
    • Based on a number of different measures. (See page 47 or look here.
    • The standards evolve over the years.
    • But the higher the number the better.