UW AMath High Performance Scientific Computing
 
Coursera Edition

Table Of Contents

Previous topic

Binary/metric prefixes for computer size, speed, etc.

Next topic

Storing information in binary

This Page

Computer Architecture

This page is currently a very brief survey of a few issues. See the links below for more details.

Not so long ago, most computers consisted of a single central processing unit (CPU) together with some a few registers (high speed storage locations for data currently being processed) and main memory, typically a hard disk from which data is moved in and out of the registers as needed (a slower operation). Computer speed, at least for doing large scale scientific computations, was generally limited by the speed of the CPU. Processor speeds are often quoted in terms of Hertz the number of machine cycles per second (or these days in terms of GigaHertz, GHz, in billions of cycles per second). A single operation like adding or multiplying two numbers might take more than one cycle, how many depends on the architecture and other factors we’ll get to later. For scientific codes, the most important metric was often how many floating point operations could be done per second, measured in flops (or Gigaflops, etc.). See Flops for more about this.

Moore’s Law

For many years computers kept getting faster primarily because the clock cycle was reduced and so the CPU was made faster. In 1965, Gordon Moore (co-founder of Intel) predicted that the transistor density (and hence the speed) of chips would double every 18 months for the forseeable future. This is know as Moore’s law This proved remarkably accurate for more than 40 years, see the graphs at [wikipedia-moores-law]. Note that doubling every 18 months means an increase by a factor of 4096 every 14 years.

Unfortunately, the days of just waiting for a faster computer in order to do larger calculations has come to an end. Two primary considerations are:

  • The limit has nearly been reached of how densely transistors can be packed and how fast a single processor can be made.
  • Even current processors can generally do computations much more quickly than sufficient quantities of data can be moved between memory and the CPU. If you are doing 1 billion meaningful multiplies per second you need to move lots of data around.

There is a hard limit imposed by the speed of light. A 2 GHz processor has a clock cycle of 0.5e-9 seconds. The speed of light is about 3e8 meters per second. So in one clock cycle information cannot possibly travel more than 0.15 meters. (A light year is a long distance but a light nanosecond is only about 1 foot.) If you’re trying to move billions of bits of information each second then you have a problem.

Another major problem is power consumption. Doubling the clock speed of a processor takes roughly 8 times as much power and also produces much more heat. By contrast, doubling the number of processors only takes twice as much power.

There are ways to continue improving computing power in the future, but they must include two things:

  • Increasing the number of cores or CPUs that are being used simultaneously (i.e., parallel computing)
  • Using memory hierachies to improve the ability to have large amounts of data available to the CPU when it needs it.

Further reading

See the Computer architecture references section of the bibliography.

See also the slides from [Yelick-UCB], [Gropp-UIUC] and other courses listed in the bibliography.