Speed of Computers
From IAE-Pedia
Contents |
Most people (myself included) have trouble comprehending bow fast computers are. From time to time I try to explain this to myself. This document is an attempt that I made in June 2006. This attempt was made shortly after I read the following article:
- Oakland Tribune (6/2806). Livermore supercomputer sets software speed record. Retrieved 3/25/09: http://findarticles.com/p/articles/mi_qn4176/is_20060628/ai_n16505665.
Quoting from the article:
- Federal weapons officials announced last week the performance record of more than 200 trillion calculations a second, partly by way of proving American dominance in simulating the workings of nuclear weaponry.
That is, this is a federally funded project in which a very fast computer is being used to simulate part of what goes on in a nuclear weapon explosion.
200 Trillion Operations a Second
200 trillion is a large number. Now, less than three years later, the top computer speeds are about five times that fast. Thus the term petaflop (1,000 trillion floating point operations per second) has come into use. See the 11/18/2008 article "Supercomputers break petaflop barrier."
The 200 trillion operation per second computer had about 131, 000 microprocessors. That is, the supercomputer can be thought of as being built by appropriately connecting about 131,000 microcomputers each running at a speed of 1.5 billion computations per second. (Note that 131,000 times 1.5 billion is approximately 200 trillion.)
However, it is quite difficult to program a multiprocessor computer so that all of its processors are being effectively used at their full potential speed. From the information provided in the article, I surmise that the processors were actually capable of a speed of 3 billion operations per second.
As noted above, the November 2008 fastest super computers were about five times as fast as the 200 trillion operation per second speed leader in the year 2006. The numbers used in the remainder of this 2006 article are out of date by a factor of five compared to the late 2008 numbers. They will become more and more out of date as super computers make use of still large numbers of microprocessors and the speed of the microprocessors grows.
My Early University of Oregon Computer Experiences
I took a job at the University of Oregon beginning in fall 1967. The UO had recently built a new building to house its new IBM 360/50. This computer was intended to serve most of the computing needs of the entire campus. It was about two or three years later when the UO added a Digital Equipment Corporation PDP-10 timeshared system.
The IBM 360/50 had a memory cycle time of 2 microseconds, and the the PDP-10 memory cycle time was 1 microsecond. A microsecond is a millionth of a second. The 360/50 was first sold in 1965, while the PDP-10 was first sold in 1967.
According to data available at http://homepage.virgin.net/roy.longbottom/cpumix.htm, the PDP-10 could average about 450,000 operations per second and the IBM 360/50 could average about 150,000 operations per second. That is, the second large computer the UO purchased was about three times as fast as the first large computer it purchased.
Just for the fun of it, I took the two pieces of data from the IBM 360/50 and PDP-10, and compared them with the number 200 trillion for the year 2006 Lawrence Livermore computer.
- IBM 360/50; vintage 1965. 200 trillion divided by 150,000 is about 1.33 billion. The year 2006 computer was about 1,330 million times as fast as an IBM 360/50.
- PDP-10; vintage 1967. 200 trillion divided by 450,000 is about .44 billion. The year 2006 computer was about 440 million times as fast as PDP-10.
In browsing for some data to use in this discussion, I saw an article that claimed it costs $150 per CPU hour to rent time on an IBM 360/50. At this charge rate per computation, not taking into consideration inflation, it would cost 200 billion dollars an hour to use the new, fast machine. If we included inflation, the cost in current dollars would be about a trillion dollars an hour.
Moore's Law
Gordon Moore was a co-founder of Intel Corporation in 1968. As he contributed to and observed progress in making more and more densely packed computer chips, he detected a pattern that has come to be called Moore's Law. The pattern was that the number of transistors incorporated into a single chip was doubling every two years.
Moore's Law is not really a "Law" in the sense of laws of science. Rather, it is a formula describing the increasing density of transistors in CPU chips, and increasing speed of CPU chips, that has proven relatively accurate over a period of 40 years. This relatively predictable rate of change has allowed people to begin to think about and develop computer systems and applications a number of years in advance of the time they would be economically viable.
While it is easy to talk about a doubling in less than two years, when this happens for decades, the numbers tend to become mind numbing. As an example, consider the UNIVAC computer that first became commercially available in 1951. The "mass production" of this machine during 1951 to 1954 produced 56 computers. During the subsequent 50 years, the price to performance of computers decreased by a factor of more than two billion! I can now buy a microcomputer for 1/2000 of the cost of the UNIVAC, and that is well over a million times as fast. Two more decades of computer performance improving at the rate it has during the past four decades would add another factor of 10,000 into the speed gain in this UNIVAC example!
Raw Speed and Actual Efficiency
Here is a short article from the April 8, 2009 ACM TechNews.
World's Most Efficient Supercomputer Gets to Work IDG News Service (04/02/09) Williams, Martyn
- The new Fujitsu FX1 supercomputer in Japan has a peak performance of 110.6 teraflops, making it the most powerful machine in Japan and the most efficient supercomputer in the world. The peak performance when running the Linpack benchmark represents 91.2 percent of its theoretical performance of 120 teraflops, and outperforms the previous record holder, a machine at the Leibniz Rechenzentrum in Munich, Germany. The Fujitsu FX1 at the JAXA's Chofu Space Center in western Tokyo has 3,008 nodes, each of which has a 4-core Sparc64 VII multiprocessor and 94 terabytes of memory. In addition to the previous, less powerful FX1 supercomputer, the facility that hosts the new FX1 also has an NEC SX-9 vector computer for specialized tasks. In total, a petabyte of disk storage space and 10 petabytes of tape storage are available. The machine will be used by Japan's Aerospace Explorations Agency to simulate the launch of spacecraft to help engineers properly insulate payloads to prevent the vibrations and noises that occur during launch from damaging satellites before they can be deployed. The new computer will help researchers capture frequencies of 150 Hz to 200 Hz, which was difficult to do on the previous computer. Fujitsu says the high efficiency of the machine is due to the high-performance hardware, the Parallelnavi middleware used in the system, and the expertise of the system's builders.
Computers Brain Versus Human Brain
Progress is occurring both in understanding how the human brain does what it does, and in developing computer programs that are accurate simulations of of the human brain. See, for example, A Working Brain Model, a November 2007 article in Technology Review.
A human brain has about 100 billion neurons. A 1,000 trillion operation per second model of a human brain could devote 10,000 operations per second to each neuron.
It turns out that this is not nearly enough compute power to adequately simulate a neuron functioning in a human brain. In a human brain, a typical neuron has about 5,000 synaptic connections to other neurons. Researchers in this field seem to have considerably varying opinions as to when the hardware, software, and knowledge of how a human brain functions might be sufficient to develop a good simulation of a human brain. Perhaps by sometime in the range of 2020 to 2025???
Note that it is not necessary to model a complete brain to produce a really valuable research tool useful in brain research. For example, here is some material quoted from a May 29, 2008 article discussing work being done at Carnegie Mellon:
- Scientists at Carnegie Mellon University have taken an important step toward understanding how the human brain codes the meanings of words by creating the first computational model that can predict the unique brain activation patterns associated with names for things that you can see, hear, feel, taste or smell. Researchers previously have shown that they can use functional magnetic resonance imaging (fMRI) to detect which areas of the brain are activated when a person thinks about a specific word. A Carnegie Mellon team has taken the next step by predicting these activation patterns for concrete nouns — things that are experienced through the senses — for which fMRI data does not yet exist.
- The work could eventually lead to the use of brain scans to identify thoughts and could have applications in the study of autism, disorders of thought such as paranoid schizophrenia, and semantic dementias such as Pick’s disease.
- …
- “We believe we have identified a number of the basic building blocks that the brain uses to represent meaning,” said Mitchell, who heads the School of Computer Science’s Machine Learning Department. “Coupled with computational methods that capture the meaning of a word by how it is used in text files, these building blocks can be assembled to predict neural activation patterns for any concrete noun. And we have found that these predictions are quite accurate for words where fMRI data is available to test them.
Educational Implications
Computer futurists think that in perhaps 20 years, individual students will have routine access to a computer with a speed of 200 trillion operations a second, or more. What type of education prepares a student to make effective use of such compute power?
Answers to this type of difficult question lie in what constitutes the essence of an academic discipline. Solving problems and accomplishing tasks lies at the very core of each discipline.
Computers bring us new ways to represent and solve problems. They empower the problem-solving and task-accomplishing people in a discipline. Over time, software gets better, hardware gets faster, and the accumulated and accessible knowledge or the human race grows. All of these ideas combine in any particular discipline to help increase the capabilities of a knowledge worker in the discipline.
Thus, the researchers, practitioners, and educators in each discipline need to examine their discipline form the point of view of how current and future computers can aid in representing and solving the problems the discipline addresses. They then need to move our education system in the direction of students learning to make effective use of such tools in each discipline they study.
References
Carnegie Mellon (5/29/08). Carnegie Mellon computer model reveals how brain represents meaning. e! Science News. Retrieved 3/36/09: http://esciencenews.com/articles/2008/05/29/carnegie.mellon.computer.model.reveals.how.brain.represents.meaning.
FLOPS - Wikipedia, the free encyclopedia.
Quoting from the article:
- In computing, FLOPS (or flops or flop/s) is an acronym meaning FLoating point Operations Per Second. The FLOPS is a measure of a computer's performance, especially in fields of scientific calculations that make heavy use of floating point calculations, similar to instructions per second. Since the final S stands for "second", conservative speakers consider "FLOPS" as both the singular and plural of the term, although the singular "FLOP" is frequently encountered. Alternatively, the singular FLOP (or flop) is used as an abbreviation for "FLoating-point OPeration", and a flop count is a count of these operations (e.g., required by a given algorithm or computer program). In this context, "flops" is simply the plural rather than a rate.
Graham-Rowe, Duncan (11/28/07). A Working Brain Model. Technology Review. Retrieved 3/26/09: http://www.technologyreview.com/Biotech/19767/.
Inflation Calculator (n.d.). Retrieved 3/26/09: http://data.bls.gov/cgi-bin/cpicalc.pl.
Longbottom, Roy (n.d.). Computer Speeds From Instruction Mixes pre-1960 to 1971. Retrieved 3/26/09: http://homepage.virgin.net/roy.longbottom/cpumix.htm.
Moore’s Law (n.d.). Gordon Moore. Accessed 3/14/09: http://www.intel.com/technology/mooreslaw/index.htm.
Supercomputer —Wikipedia, the free encyclopedia.
Quoting from this article:
- A supercomputer is a computer that is at the frontline of current processing capacity, particularly speed of calculation. Supercomputers introduced in the 1960s were designed primarily by Seymour Cray at Control Data Corporation (CDC), and led the market into the 1970s until Cray left to form his own company, Cray Research. He then took over the supercomputer market with his new designs, holding the top spot in supercomputing for five years (1985–1990).
Oakland Tribune (6/28/06). Livermore supercomputer sets software speed record. retrieved 3/26/09: http://findarticles.com/p/articles/mi_qn4176/is_20060628/ai_n16505665?tag=content;col1.
Top 500 Computer Sites. Retrieved 3/25/09: http://www.top500.org/.
- This Website contains a list of the 500 fasted computers in the worlds, and the list is updated about twice a year. The Website also contains information about the supercomputer industry.
Author or Authors
The original version of this document was written by David Moursund.
