Supercomputers

Preface

This document is dedicated to all the men and women who are involved with ensuring the national security of the United States.

The following information was derived from these sources:

Body of Secrets by James Bamford.
Wizards of Langley by Jeffrey T. Richelson.
The Bureau by Ronald Kessler.
Wikipedia, the free encyclopedia.
Aviation Week & Space Technology, November 27, 2006, page 22.

Although I have attempted to gather and arrange the data into a comprehensive, understandable format, any errors and omissions are mine and not attributable to any other authors or works. This document was created prior to the reorganization of the intelligence community following 9/11, so in many respects, it is out-of-date.

This paper is designed to provide basic information to persons wishing to enter the US Intelligence Community and those who desire a broader knowledge of it. This paper is not intended to cover every aspect of the intel community.

Supercomputers

2 x 10 to 6............Cray 1 (Lodestone).............(2 megaflops[1]).......1976
100 x 10 to 6..........Cray X-MP......................(100 megaflops).....1983
1.2 x 10 to 9..........Cray 2 (Bubbles)...............(1.2 gigaflops).....1985
4 x 10 to 9............Cray X-MP (enhanced)...........(4 gigaflops).......1988
6.5 x 10 to 9..........Thinking Machines Corp.[4].....(6.5 gigaflops).....1991
20 x 10 to 9...........NEC SX-4 (for NCAR[5]).........(20 gigaflops)......1996
20 x 10 to 9...........IBM Deep Computing.............(20 gigaflops)......2001
1.3 x 10 to 12.........DOD TSE-1200...................(1.3 teraflops).....2001
20 x 10 to 12..........Cray SV-2 (SGI/TERA)...........(20 teraflops)......2002
1 x 10 to 15...........RIKEN MDGRAPE-3[7].............(1 petaflops).......2006
596 x 10 to 12.........IBM Blue Gene/L................(596 teraflops).....2007
101 x 10 to 12.........Cray XT/4......................(101 teraflops).....2007
839 x 10 to 12.........NEC SX-9.......................(839 teraflops).....2008
500 x 10 to 12.........Sun Ranger[6]..................(500 teraflops).....2008
1 x 10 to 15...........IBM Blue Gene/P................(1 petaflops).......2008
1 x 10 to 15...........NASA, CGI, Intel...............(1 petaflops).......2009
2 x 10 to 15...........DARPA[2] HPCS[3]...............(2 petaflops).......2010
10 x 10 to 15..........NASA, CGI, Intel...............(10 petaflops)......2012

Notes:
[1] FLOPS is an acronym relating to computer speed and stands for FLoating point Operations Per Second (FLOPS), similar in concept to Instructions Per Second (IPS).
[2] DARPA is the Defense Advanced Research Projects Agency (DARPA).
[3] HPCS is High Productivity Computing Systems (HPCS) program, an ongoing project run by DARPA to advance the speed of computers. The online date for the first HPCS, being built by IBM and Cray, Inc., is 2010. When these fast computers become available, they might be used for the following tasks:

Ship and aircraft design

Stealth design

Weapons simulations

Explosives simulations

Code-breaking

U.S. nuclear stockpile analysis

Weather Simulation

HPCS computers will probably be made available to the following entities:

NSA

DOE Office of Science

National Nuclear Security Agency (NNSA)

[4] Thinking Machines Corp Connection Machine 5
[5] NCAR is an acronym for the National Climate and Atmospheric Research Agency
[6] The Sun Ranger was built in conjuction with the National Science Foundation (NSF), American Micro Devices (AMD), and the University of Texas for open science research.
[7] RIKEN is the Japanese research institute. This is not a general purpose computer as are the others in the list. It was specially built to simulate molecular dynamics.

US Numbering Table

1 x 10 to the 0 power: one - flops

1 x 10 to the 3 power: thousand - kiloflops

1 x 10 to the 6 power: million - megaflops

1 x 10 to the 9 power: billion - gigaflops

1 x 10 to the 12 power: trillion - teraflops

1 x 10 to the 15 power: quadrillion - petaflops

1 x 10 to the 18 power: quintillion - exaflops

1 x 10 to the 21 power: sextillion - zettaflops

1 x 10 to the 24 power: septillion - yottaflops

1 x 10 to the 27 power: octillion

1 x 10 to the 30 power: nonillion

1 x 10 to the 33 power: decillion

1 x 10 to the 36 power: undecillion

1 x 10 to the 39 power: duodecillion

1 x 10 to the 42 power: tredecillion

1 x 10 to the 45 power: quattuordecillion

1 x 10 to the 48 power: quindecillion

1 x 10 to the 51 power: sexdecillion

1 x 10 to the 54 power: septendecillion

1 x 10 to the 57 power: octodecillion

1 x 10 to the 60 power: novemdecillion

1 x 10 to the 63 power: vigintillion

1 x 10 to the 303 power: centillion

Distributed computing Groups

Distributed Computing uses the Internet to link personal computers to achieve high speed processing.

BOINC - averages over 1 petaflops as of March 16, 2008.

SETI@Home - computes data averages at more than 265 teraflops.

Folding@Home - has reached over 1 petaflops as of September 15, 2007. Note, as of March 22, 2007, PlayStation 3 owners may now participate in the Folding@home project. Because of this and high performance GPU clients, Folding@home is now sustaining over 2 petaflops as of May 8, 2008.

Einstein@Home is crunching more than 100 teraflops.

As of June 2007, GIMPS is sustaining 23 teraflops.

Processor speeds

Intel Corporation recently unveiled the experimental multi-core POLARIS chip, which achieves 1 teraflops at 3.2 GHz. The 80-core chip can increase this to 1.8 teraflops at 5.6 GHz, although the thermal dissipation at this frequency exceeds 260 watts.
As of 2007, the fastest PC processors (quad-core) perform at more than 30 gigaflops. General purpose Processing units (GPUs) in PCs are considerably more powerful in pure flops. For example, in the GeForce 8 Series of PCs, the nVidia 8800 Ultra performs around 576 gigaflops on 128 Processing elements. This equates to around 4.5 gigaflops per element, compared with 2.75 per core for the Blue Gene/L. It should be noted that the 8800 series performs only Single precision calculations, and that while GPUs are highly efficient at calculations, they are not as flexible as a general purpose CPU (GPU).

Cost of computer hardware for 1 gigaflops

1961: about US$1,100,000,000,000 ($1.1 trillion) per 1 gigaflops (=US$1,100 per flops); with 1 billion IBM 1620 units @ $64,000 each and a multiplication operation taking 17.7ms.

1997: about US$30,000 per 1 gigaflops; with two 16-Pentium-Pro�processor Beowulf cluster computers.

2000, April: $1,000 per GFLOPS, Bunyip, Australian National University. First sub-US$1/megaflop. It won the Gordon Bell Prize 2000.

2000, May: $640 per 1 gigaflops, KLAT2, University of Kentucky.

2003, August: $82 per 1 gigaflops, KASY0, University of Kentucky.

2006, February: about $1 per 1 gigaflops in ATI PC add-in graphics card (X1900 architecture) � these figures are disputed as they refer to highly parallelized GPU power.

2007, March: about $0.42 per 1 gigaflops in Ambric AM2045.

2007, October: about $0.20 per 1 gigaflops with the cheapest retail Sony PS3 console, at US$400, that runs at a claimed 2 teraflops. These figures represent the processing power of the GPU. The seven CPUs run collectively at a lower 218 gigaflops.

Moore's Law

The trend toward lower and lower cost for the same computing power follows Moore's law. Moore's original statement that transistor counts had doubled every year can be found in his publication "Cramming more components onto integrated circuits", Electronics Magazine 19 April 1965:

"The complexity for minimum component costs has increased at a rate of roughly a factor of two per year ... Certainly over the short term this rate can be expected to continue, if not to increase. Over the longer term, the rate of increase is a bit more uncertain, although there is no reason to believe it will not remain nearly constant for at least 10 years. That means by 1975, the number of components per integrated circuit for minimum cost will be 65,000. I believe that such a large circuit can be built on a single wafer."

The term "Moore's Law" was coined around 1970 by the Caltech professor, VLSI pioneer, and entrepreneur Carver Mead. Moore may have heard Douglas Engelbart, a co-inventor of today's mechanical computer mouse, discuss the projected downscaling of integrated circuit size in a 1960 lecture.

In 1975, Moore altered his projection to a doubling every two years. Despite popular misconception, he is adamant that he did not predict a doubling "every 18 months." However, an Intel colleague had factored in the increasing performance of transistors to conclude that integrated circuits would double in performance every 18 months.

In April 2005, Intel offered $10,000 to purchase a copy of the original Electronics Magazine dated 19 April 1965. David Clark, an engineer living in the UK, was the first to find a copy and offer it to Intel.

Wirth's law

"Software gets slower faster than hardware gets faster".