Computation History & Technology

Dr. John F. Jarvis

USC Aiken, Dept. of Mathematical Sciences, retired

I have been presenting an exhibit at the University of South Carolina-Aiken annual Science Education Enrichment Day “SEED” that has been designed to convey two important facts about computers. First, the idea or concept of what a computer is was established before the first computers were created. Second, even though the technology used to build computers has improved immensely over the past 60+ years the basic concept of the computer hasn't changed. This page describes the exhibit.

The central intellectual idea in computing is the algorithm, a list of well defined steps performed by a computer or person to accomplish a specific task. These steps are defined in terms of the task to be accomplished. As an example, consider solving Rubik's Cube. The steps that are used by a solution algorithm are the 18 possible quarter and half turn twists of the six faces of the cube. A solution of a scrambled cube can be completely specified as a list composed from these 18 basic steps. A solution algorithm starts with the colors of each facet on each face of the cube. Sequences of face twists are progressively applied to solve the cube. A person can do this directly or they could be used as the basis of a computer program to do the same task. The computer generated solution sequence could be demonstrated on a simulated cube (graphical display) or used to instruct a robot to solve an actual Rubik's cube. Computer programming is the conversion of an algorithm specified in task related steps to a list computer instructions that will perform the same task on a representation of the problem. In the case of the cube, a sufficient representation is a list of 54 numbers specifying the color of the facets. The most used explanation of what constitutes a satisfactory algorithm and a minimal computer for processing it was given by Alan Turing in 1936. (The Advent of the Algorithm, David Berlinski, Harcourt 2000)

The 21 year period starting with World War II (1940-1960) saw the birth of the computer, invention of the transistor (1947) and the invention of the integrated circuit (1958). The period starts with calculations done on manually operated mechanical calculators and ends with the availability of reliable, but expensive, semiconductor based main frame computers. The last half of this interval includes the first decade of commercially sold computers. The 45+ years since then have been building on this foundation with performance approximately doubling every one and a half to two years (Moore's Law).

Computing progress is displayed in the following chart that positions each system or computation method at the logarithm of its numerical computation speed. Each tick mark represents a doubling of the computation rate. This quantity is a more realistic indicator of the usefulness of a computer system than its actual computing speed. Determination of FLOPS (FLoating decimal point Operations per Second) is difficult as it depends on the arithmetic operation being done and on the size of the operands. I have used the time for floating point multiplies where I have been able to determine it. Fortunately, errors of 2x or more won't significantly change this plot. Each system is summarized by it's name, technology, type of system, introduction date, floating point calculation rate in FLOPS and approximate system cost.

The geometric mean of the processing capabilities, midpoint on the logarithmic scale, corresponds to the IBM 709/7090 computers. Consider this additional evidence of the maturity of late 50s or early 60s computing.

FLOPS - FLoating point OPerations per Second
M - Mega (millions)
G - Giga (billions)

By 1971 IC technology had progressed sufficiently to enable INTEL to design and fabricate the first single chip CPU, the 4004.

1985 saw the introduction of the INTEL 386 CPU chip, perhaps the first CPU chip containing all features required to run modern software. Newer CPU chips simply offer additional performance.

Not included in the chart, but equally important, are the reductions in power and size, and the gain in reliability obtained as semiconductor technology matures.

Computation speed is not the only measure of computing technology. Random Access Memory (RAM), disk storage and data transmission technologies have developed to keep pace with the computation technologies.

System costs are given at the time of introduction, not in present dollar values. Obviously, this makes the older systems even more expensive when compared to today. Technology development provides both faster and cheaper systems.

Semiconductor technology progresses by making the size of the individual parts of a circuit (transistors and wiring) smaller. This is a win in two ways. More parts can be put on a single chip and the transistors switch faster. The small size of the parts on a integrated circuit make exhibiting this progress difficult. Printed Wiring Board (PWB) technology shows a similar but much less dramatic improvement as its technology develops. Fortunately, the scale of PWB makes it easy to see the improvements.

This exhibit presents examples of the type of technologies used in each step of the development in the form of calculators and simple counting circuits. The relay counter uses an electromechanical device that directly counts in decimal. Each electronic counter consists of four flip-flops connected such that transition from on to off causes the next flip-flop to change. Flip-flops store a single bit of memory and are used to implement static memory in computer systems. Compare the different examples for speed, power and number of components required to perform the same function. For all the electronic counters the count value of each stage is the binary sequence 1 - 2 - 4 - 8 reading from left to right and is shown on the small labels adjacent to each lamp or LED. The decimal value is obtained by summing the values for the lit LEDs or neon bulbs.

It should be evident from the counters and calculators exhibited that the ideas and methods of computation don't depend on how these computations are implemented. The essential nature of computing was understood before World War II and hasn't changed. Improving technology makes faster and cheaper computation systems available but does not change the nature of the computations.

A much more detailed article describing the history of computer hardware is available at the Wikipedea.

Present desktop computers (2006) are approximately 1,000,000 (million) times faster and 1,000 times cheaper that the vacuum tubed IBM 709, an improvement of a billion (10⁹, 1,000,000,000). If the system price was kept constant the factor of a billion could all be in speed. Such systems are created and are called "super computers". To put the factor of a billion in perspective, a billion seconds is almost 32 years.

ABACUS

Used:

earliest - before Christian era, still in use in many parts of Asia

Type:

decimal digital, primarily addition/subtraction, algorithms (methods) for multiplication and division exist.

Example:

Home built Chinese style, 10 decimal digits (columns of beads)
Accuracy is determined by the number of digits (columns)
This style dates from the 14th century

This image shows a ten column abacus displaying the values 0 - 9 reading from left to right. The upper frame consists of two beads per column each having a value of five. The zero value position is against the top edge. Each bead in the lower frame is worth one and the zero position is against the bottom edge.

The Universal History of Numbers, Georges Ifrah, Wiley, 2000
The Abacus, Jesse Dilson, St. Martin's Press, 1994 (reprint of 1968 ed.)

http://www.ee.ryerson.ca:8080/~elf/abacus/

SLIDE RULE

Used:

Invented by William Oughtred (1574-1660) in 1630.
Continued in use until advent of electronic pocket calculators in early 1970s.

Description:

Analog (lengths proportional to logarithms)
Performs multiplication, division and function value determination.
Logarithms convert multiplication to addition and division to subtraction.
Accuracy is limited by precision of manufacture and operator skill.
Results to 3 decimal digits would be excellent.
Decimal point position is the responsibility of the operator.

Examples: Precision Sterling - a student instrument of unknown age
K&E LOG LOG DUPLEX DECITRIG - I purchased this one in 1957 during my sophomore year in high school and used it until 1968.

Image: top - K&E Log Log Decitrig, bottom - Precision Sterling
Image - close up of K&E cursor. The logarithmic nature of scales is evident

http://en.wikipedia.org/wiki/Slide_Rule
http://www.oughtred.org/
http://www.sphere.bc.ca/test/sruniverse.html
http://www-gap.dcs.st-and.ac.uk/~history/Mathematicians/Oughtred.html

MECHANICAL CALCULATOR

Used: mid 17th century to early 1970s

In large scale calculations, average rates of 1 floating point operation per minute were typical. Large scale calculations involve the combined effort of many operators. Checking of work in progress, inevitable breaks, recording and coordination of results gives average computation rates somewhat less than the time estimated from a small number of calculations.

The word computer originally meant the operator of one of these calculators.

Prof. David Alan Grier provided an estimate of 0.25 - 10 operations/minute for manually operated mechanical calculators. The higher speeds are only for addition.

Examples:

Lightning Calculator
Only addition and subtraction
Manufactured 1900s - 1960s

Monroe SN-212
Addition, Subtraction, Division (Multiplication by shift and repeated add)
Manufactured around 1960

Image: Lightning calculator and stylus Addition is accomplished by inserting the stylus in the desired number (large & white) and turning the dial clockwise until it stops. Subtraction is performed using the smaller, yellow numbers turning the dial in a counter clockwise direction.

Image: Monroe SN212.

Calculator with this technology:

Babbage Difference Engine
Proposed 1821
Built 1991 by the Science Museum of London

Computer with this technology:

Babbage Analytic Engine
Proposed 1836

"The Difference Engine. Charles Babbage and the Quest to Build the First Computer", Doron Swade, Viking, 2000

http://www.vintagecalculators.com/html/mechanical_calculators.html
http://en.wikipedia.org/wiki/Babbage

RELAYS

What:

Electrically operated mechanical switches
Developed by ATT for dial telephone exchanges used until early 1960s
Operation is slow because of the mechanical motions required
Counting rate: 1-10 counts/second
For computing this was a technological dead end

Counter:

A “stepping relay” is used to implement a decade counter
Each time the relay coil is actuated, the stepper advances one increment
Small incandescent lights are used to indicate the value
Relay requires 10W during actuation but no power otherwise
Bulbs require 0.75W
Counting rate: 10 Hz

Image of the Relay Counter

Calculator with this technology:

Bell Lab Complex Number Calculator
research project, first of several relay based computers
operational 1939
very reliable
was accessed remotely using a teletypewriter
1 FLOPS
cost: about $20,000

Chapter 9 in "A History of Engineering & Science in the Bell System: Communication Sciences (1925-1980)", Ed. S. Millman, AT&T Bell Laboratories, 1984

VACUUM TUBE BINARY COUNTER

Counter:

Power : 14 Watts
Counting rate: 100 KHz
When: 1930 - 1950's

This counter was built in 2002. Each vacuum tube actually contains two triodes (amplifiers). Each counter stage (flip-flop) requires two amplifiers and stores 1 bit of information. Four flip-flops enable counting to 16. The vacuum tubes, neon lamps and power transformer date from the 50's or 60's. The small components are new. This counter is typical of 1950's electronics. The wiring can be viewed through the plastic window on the front of the counter. The protective window is necessary because the circuits require 200 volts DC to operate.

Image of the vacuum tube counter

Image showing vacuum tube counter wiring

Computers with this technology:

ENIAC (Electronic Numeric Integrator, Analyzer and Computer)
Research
Operational 1945
385 FLOPS (floating point operations/second)
Main memory consisted of 20 registers containing 10 decimal digits
Used 18000 vacuum tubes, almost 200 KW power
Approximate cost $500,000

IBM 709 "main frame"
Introduced 1957
5000 FLOPS
Used 6600 vacuum tubes
32768 36 bit word magnetic core main memory
Approximate cost: $2,000,000 up

The straight sided, plastic base 6SN7 is also a dual triode and is typical of 1940's electronics. The even larger "light bulb" style of tubes was common in the 1930's. These older tubes often require up to twice as much power to perform the same function as the tubes in this counter. The ENIAC uses the 6SN7 and older style tubes. The IBM 709 used tubes similar to those in the counter.

Image of older tubes and a 1950s commercial decade counter. Note the use of a primitive PWB

Besides high power consumption, vacuum tubes are quite unreliable. The average time to failure for large systems based on tubes is measured in hours.

I started my programming career on an IBM 709 during my senior year (1962) at the University of Florida.

"Bit by Bit: An Illustrated History of Computers", Stan Augarten,Ticknor & Fields, New York, 1984
"Reference Manual 709 Data Processing System", IBM, ©1959-1961

http://ftp.arl.mil/~mike/comphist/eniac-story.html
http://www.eingang.org/Lecture/eniac.html
http://www.columbia.edu/acis/history/ibm709.html

DISCRETE TRANSISTOR BINARY COUNTER

Counter:

Power: 0.10 W
Counting rate: 1 MHz
Typical of 1960's

This circuit was built in 2002 using parts, other than the LEDs, that were available in the 60's. Each counter stage requires two transistors (amplifiers). This first generation of solid state circuitry simply replaces vacuum tubes with transistors.

Commercial products would use printed wiring boards (PWB), not the point-point wiring used in these examples. PWBs became common in the late 1950s.

Image of all three semiconductor counters and the 1974 calculator. For all the electronic counters the count value of each stage is the binary sequence 1 - 2 - 4 - 8 reading from left to right. The decimal value is obtained by summing the values for the lit LEDs.

Computer with this technology:

IBM 7090 "mainframe" (transistorized IBM 709)
Introduced: 1960
25000 FLOPS
Approximate cost: $2,000,000

http://www.columbia.edu/acis/history/7090.html

SSI (Small Scale Integrated Circuit) COUNTERS

Counters:

Power: Binary: 0.085 W (no LEDs on) - 0.185 W ( all LEDs on)
Decimal: .205 W ("1") - 0.290 W ("8")

Counting rate: 10 MHz
When 1970's

These circuits were built in 2002 using parts and technology (TTL) that became available in the 70's.

A single Silicon Integrated Circuit (SIC or just IC) implements the four flip-flops in each counter. A second IC contains six logic inverter circuits. Four are used to drive the LEDs. A fifth one is used to implement a reliable interface to the push button. All the resistors are used as part of the display circuitry and the push button interface. One capacitor is used as part of the push button interface and the second is required to protect the counters from electrical noise.

The capability of adding logic to the same number of packages is shown in the decade counter. In this counter extra logic is added to the counter chip making it a decade counter. The third IC is a driver for the 7 segment, each a LED, display and contains the logic needed to convert the binary value to the segments that must be lit to display the appropriate decimal number. The 7 segment driver has 16 leads (electrical connections), 2 more than the other ICs in the counters.

Each of these ICs contain 100s of transistors and resistors.

Closeup of the solid state decimal counter

Computer with this technology:

DEC 11/780 "mini computer"
Introduced: 1978
500000 FLOPS
Approximate cost: $250,000

The research group that I was a part of at Bell Labs had an early model of this system.

http://en.wikipedia.org/wiki/PDP-11

http://ed-thelen.org/comp-hist/VAX-11-750.html

ELECTRONIC CALCULATOR

Example:

Texas Instruments Exactra 20
"four function" - add/subtract/multiply/divide
Provides (only) 6 digit LED display with sign indicator
Decimal point must be within displayed digits
Power: 0.45 - 0.50 W (Depends on number of digits displayed)
Short battery life
Introduced: 1974

The logic is implement by a TMS0135 SIC, a 7439 SIC and 30 discrete parts. The earlier TMS0111 which was also designed for 4 function calculators had approximately 20000 transistors. (see http://smithsonianchips.si.edu/ Item NMAH Catalog Number 1987.0487.233)

Image of all three semiconductor counters and the 1974 calculator

Recently (2001), I purchased a new generic 56 function, 12 digit display scientific calculator for $10 at a local grocery store. The power consumption of this device is 50 micro watts, about 1/10000 of the Exactra 20. Technology advance provides for more than just performance increases.

http://www.hpmuseum.org/ (includes history preceding electronic era)

LSI (Large Scale Integrated Circuit) COMPUTERS

LSI is building a complete system or subsystem as a single integrated circuit. The number of transistors that can be fabricated on a single chip has increased exponentially (Moore's Law) consequently the definition of LSI continually increases. The TMS 0135 calculator chip would qualify as LSI in 1974. Good examples of 2002 LSI are the INTEL Pentium 4 CPU that contains about 55 million transistors and the Radeon R300 graphics processor with approximately 110 million transistors.

First Single Chip CPU:

INTEL 4004
Introduced 1971
Instruction rate: 0.108 MHz, perhaps 100 FLOPS (Is compared to the ENIAC is some descriptions. Use 7 for the log2(FLOPS) value to place on the performance plot.)
2300 transistors
Packaged in a 16 Pin DIP, Dual Inline Package, same as the 7 Segment Display Driver
Initially used to implement a printing calculator, price unknown

Breakthrough LSI CPU chip set:

INTEL 386/387
Introduced 1985
Includes all functions needed to support modern software
Floating point calculations are done in the 387 coprocessor
Clock 16 MHz, 2 million FLOPS
275,000 transistors in the 386 , 120,000 in 387
Computer system cost: $1500 - $2000
http://en.wikipedia.org/wiki/Intel_80386

Current Single Chip CPUs:

Used in current PCs
Processors - The current rage in processors is multiple cores, more than one independent CPU on a single chip
Greater than 5 billion FLOPS.
Approximate system cost: $1000

http://en.wikipedia.org/wiki/PowerPC_970
http://www.intel.com/pressroom/kits/quickreffam.htm

 Continuing progress (Moore's Law) assures the current chip will always be out of date.

Copyright 2003-2008 John F. Jarvis