DOCSLIB.ORG

The Illiac IV System

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
The Illiac IV System ~ROCFEDINOS OF THE IEEE, VOL. 60, NO. 4, APRIL 1972 369 49, p. 439, 1966. 'H E. W. Wagner, British Patent No. 1,065796 1967. W. H. F. Talbot, "Photogem Drawing," The Athenaeum, No. 589, p. 114, 1839. lL9 L. E Walkup, US. Patent No. 2,777,957, 1957. I59 K. Tamarlbuchl.and M. L. Smlth."Charge Determiling Species in Non-Aqueous ''O L. E. Walkup, U.S. Patent No. 2.825.614. 1954. Solvents." J. Colloid Interlace Sci., VoI. 22, p. 404, 1966. P. J. Warter, Jr.. ' FactorsDeterminlng Xerographlc Photoreceptor Performance," '60 T. Taniand S. Klkuchi."Spectral Sensitlzarion in Photographyand Electrophotog- Appl. Optics: Suppl. =3 on Electrophotography. p 65, 1969. raphy," Report lnst. Industrial SCI., Univ. Tokyo, Vol. 18, p. 51, 1968. H.Watanabe. et al."The Activation Energy !or OxygenDesorptlon from Zinc Oxide A. Tereninand I. Akimov."Some Experiments on thePhotosensitization MechaTism Surfaces." Jap. J. Appl. Phys., Vol. 4, p. 945. 1965. of Semiconductors by Dyes," J. Phys. Chem., VoI 69. p. 730, 1965. "' J. W. Weigl. Photographic Science, Ed. by W. F. Berg, New York: Focal Press, 1963. IiZ V. D Tughan and R. C. Pink, "Solutions of Metal Soaps in Organic Solvents Part 11," 'I4 R. D. Weiss, "Electrolytic Photography," Phot. Sci. and Enq., Vol. 11. p. 287, 1967. J. Chem. p. 1604, 1951 Soc., 'I5 P. H. Wersema,et al, "Calculatlon of the Electrophoretic Mobility of aSpherical ' h3 V. Tuiagin."Imaging Method Based on Photoelectrophoresis," J. Opt Soc. Amer., Collold Particle," J. CoUoid Interface Sci., Vol. 22. p. 78. 1966. VoI. 59, p. 328. 1969. 'leH. Wiellcki. private communicatlon. ld4 E. J Verwey and J. Th. G. Overbeek, The- of theStability of LyophobicColloids, Amsterdam: Elsevler, 1946. 117 N. E. Wolff. "A PhotoconductiveThermoplastic Recording System," RCA Review, Vol. 25, p. 200. 1964. J. Viscakas.et al. "Recombination in Selenium Electrophotographic Layers," Awl. Optics: Suppl. -3 on Electrophotography, p. 27, 1969. 'leW. Yellin, et al, British Patelt No. 1,016,072, 1966. W. C. York, U.S. Patent No. 3,135.695. 1964. Ibb J. Viscakasand V. Gaidelis."The Role of IntercrystallineBarriers in ZnO Electro- photographicLayers," Reprographle 11, llnd Int.Conf. Cologne, Helwich Darmstadt, 'do C. J. Youngand H. G. Grieg."Electrofax: Direct Electrophotographic Printing on 1969. Paper," RCA Review, Vol. 15. p, 469, 1954. Ib7 0. Von Bronk, British Patent No. 188.030, 1922 '8' Anon., Electronic News, p. 68, February 16, 1970. The Illiac IV System W. J. BOUKNIGHT, STEWART A. DENENBERG, DAVID E. McINTYRE, J. M. RANDALL, AMED H. SmH,AND DANIEL L. SLOTNICK, SENIOR MEMBER, IEEE Invited Paper -Tho masons for the creation of llliac IV an described and Wn historyof tho liiiac IV pt+ct is mwntod. The archihchrn or hard- ware struetun of * llliac IV is discuss- llllac IV array is an army processor with a rp.dalied controlunit (CUI that urn be vkwed as a small stand-alone computer. The llliac IV sohare strategy is described in krms of current user habits and needs. Brief descriptions are given of the systems software itself, its history, and the maior lessons leaddur- ingits development. Some ideas forfuture development are suggested. Applications of lliiac N are discussed in terms of evaluating the fundon ffxJ rimultanoously on up to 64 distinct argument sets x<.Many of the time- consuming probkms In scientiAc computation invoke repeated evaludon of the same function on dihrenl agument soh. The agumenl sets which compose the problem data base must be structured in such a fashion tho+ theycan bo dimtbuted omong 64 separatememories. Two matrix ap plications: Jacobi's algorithm for Anding the eigenvalues and eigenvectors MEMORY of mal symmetric matrices, and reducing a mal nonsymmetric matrix to Fig. 1. Functional relationsa within a conventional computer. CU the uppor-Hawnberg form usingHouseholder's transformotions are dis- The cussed in detail. The ARPA networlr, a highly sophisticated and wide has thefunction of fetchinginstructions which arestored in memory, ranging experiment in the remote access and sharingof computer n- decoding or interpreting these instndons, and linally generating SOUKOS, is brieAydescribed and itscurrent status discussed. Many re- the microsequences of electronic pulses which cause the instruction marchers located about thecountry who will use Illiac IV in solving to be performed. The performance of the instruction may entail the problems will do so via the network. Thevarious systems, hardware, use or "driving" of oneof the three other components. TheCU may also a small registers can and procedures they will use is discussed. contain amount of memory called that be accessed faster than the main memory. The ALU contains the elec- I. INTRODUCI-ION tronic circuitry necess~~yto perform arithmetic and logical opera- tions. The ALU may also contain register storage. Memory is the T ALL BEGAN in the early 1950's shortly after EDVAC [l] medium by which information (instructions or data) is stored. The became operational. Hundreds, then thousands of compu- 1/0accepts information which is input to or output from Memory. 1ters were manufactured, and they were generally organized The 1/0 hardware may also take care of converting the information from one coding scheme to another. The CU andALU taken on Von Neumann's concepts, as shown and described in Fig. 1. together are sometimes called a CPU. In the decade betwetm 1950 and 190, memories became cheaper and faster, andthe conceptof archival storage was evolved; con- trol-and-arithmetic and logic units became more sophisticated: 1/0 devices expanded from typewriter to magnetic tape units, disks, drums, and remote terminals. But the four basic compo- nents of a conventional computer (control unit (0,arithmetic- Manuscript received December 10, 1971; revised January 17, 1972. and-logic unit (ALLJ), memory, and I/O) were all present in one Ttris inoitedpper is one of a series phned on topics of general interest- The Mtor. form or another. The authors are with the Center for Advanced Computation andthe The turning away fromthe conventional organization came in Illiac N Project, University of Illinois, Urbana, Ill. 61801. the middle lm's, when the law of diminishing returns began to Authorized licensed use limited to: The University of Auckland. Downloaded on May 27,2010 at 01:22:55 UTC from IEEE Xplore. Restrictions apply. 370 PR~INGSOF THE IEEE, APRIL 1972 Timing Cycle 1 2 3 4 5 Fig. 2. Pipelined operation. The large boxes representthe circuits required to transform the operands A and B into the quantity O(A, B) (some function of A and B, say the sum of A and B). The smaller boxes represent storage stages for the intermediate results O(A, B), and @A, B)2, and the desired resultO(A, B). The operation0 has been broken.down intothree stages, each of which acceptsas input the output of the pre- vious stage, and all of which perform a stage of the operation at the same time. At esch step of the timing cycle, the pipeline accepts a new pair of operands (A, B) and the previous pair moves to the next stage. This mode of operation causes results (the sum in this example) to appear at the end of the pipeline at time intervals equal to the time of operation of the slowest stage of the pipeline. take effectin the effort to increase the operational speedof a com- The replication philosophy is exemplified by the general multi- puter. Up until this point the approach was simply to speed up processor which replicates three of the four major components the operationof the electronic circuitry which comprised the four (all but the I/O)many times.The cost of a general multiprocessor major functional components. (See Fig. 1.) is, however, very high and further design options were considered Electronic circuits are ultimatelylimited in their speed of which would decrease the cost without seriously degrading the operation by the speed of light (light travels about one foot in a power or efficiency of the system. The options consist merely of nanosecond) and many of the circuits were already operating in recentralizing one of the three major componentswhich had been the nanosecond time range. So, although faster circuits could be previously replicated inthe general multiprocessor-the memory, made, the amount of money necessary to produce these faster the ALU, or the CU. Centralizing the CU gives rise to the basic circuits was not justifiable in terms of the small percentage in- organization of a vector or array processor such as Illiac IV. This crease of speed. particular option was chosen for two main reasons. At this stage of the problem two new approaches evolved. 1) Cost: A very high percentage of the cost within a digital 1) Owrlup: The hardware structure of the conventional or- computer is associated with CU circuitry. Replication of this ganization was modified so that two or more of the major func- component is particularly expensive, and therefore centralizing tional components (or subcomponents within a major compo- the CU saves more moneythan can be saved by centralizing either nent) could overlap their operations. Overlap means that more of the other twocomponents. than one operation is occurring during the same time interval, 2) Structure: There is a large class of both scientific and and thus total operation timeis decreased. business problemsthat can be solved by a computer with oneCU Before operations could be overlapped, control sequences be- (one instruction stream) and many ALUs.The same algorithm is tween the components had to be decoupled. Certainly the CU performed repetitively on mzny sets of different data: the data could at least be fetching the next instruction while the ALU was are structured as a vector, and the vector processor of Illiac IV executing the present one.
Load more

Recommended publications
  • Early Stored Program Computers
    Stored Program Computers Thomas J. Bergin Computing History Museum American University 7/9/2012 1 Early Thoughts about Stored Programming • January 1944 Moore School team thinks of better ways to do things; leverages delay line memories from War research • September 1944 John von Neumann visits project – Goldstine’s meeting at Aberdeen Train Station • October 1944 Army extends the ENIAC contract research on EDVAC stored-program concept • Spring 1945 ENIAC working well • June 1945 First Draft of a Report on the EDVAC 7/9/2012 2 First Draft Report (June 1945) • John von Neumann prepares (?) a report on the EDVAC which identifies how the machine could be programmed (unfinished very rough draft) – academic: publish for the good of science – engineers: patents, patents, patents • von Neumann never repudiates the myth that he wrote it; most members of the ENIAC team contribute ideas; Goldstine note about “bashing” summer7/9/2012 letters together 3 • 1.0 Definitions – The considerations which follow deal with the structure of a very high speed automatic digital computing system, and in particular with its logical control…. – The instructions which govern this operation must be given to the device in absolutely exhaustive detail. They include all numerical information which is required to solve the problem…. – Once these instructions are given to the device, it must be be able to carry them out completely and without any need for further intelligent human intervention…. • 2.0 Main Subdivision of the System – First: since the device is a computor, it will have to perform the elementary operations of arithmetics…. – Second: the logical control of the device is the proper sequencing of its operations (by…a control organ.
    [Show full text]
  • Computer Development at the National Bureau of Standards
    Computer Development at the National Bureau of Standards The first fully operational stored-program electronic individual computations could be performed at elec- computer in the United States was built at the National tronic speed, but the instructions (the program) that Bureau of Standards. The Standards Electronic drove these computations could not be modified and Automatic Computer (SEAC) [1] (Fig. 1.) began useful sequenced at the same electronic speed as the computa- computation in May 1950. The stored program was held tions. Other early computers in academia, government, in the machine’s internal memory where the machine and industry, such as Edvac, Ordvac, Illiac I, the Von itself could modify it for successive stages of a compu- Neumann IAS computer, and the IBM 701, would not tation. This made a dramatic increase in the power of begin productive operation until 1952. programming. In 1947, the U.S. Bureau of the Census, in coopera- Although originally intended as an “interim” com- tion with the Departments of the Army and Air Force, puter, SEAC had a productive life of 14 years, with a decided to contract with Eckert and Mauchly, who had profound effect on all U.S. Government computing, the developed the ENIAC, to create a computer that extension of the use of computers into previously would have an internally stored program which unknown applications, and the further development of could be run at electronic speeds. Because of a demon- computing machines in industry and academia. strated competence in designing electronic components, In earlier developments, the possibility of doing the National Bureau of Standards was asked to be electronic computation had been demonstrated by technical monitor of the contract that was issued, Atanasoff at Iowa State University in 1937.
    [Show full text]
  • Performing the Shuffle with the PM2I and Illiac SIMD Interconnection Networks
    Performing the Shuffle with the PM2I and Illiac SIMD Interconnection Networks Robert R. Seban Howard Jay Siegel Purdue University School of Electrical Engineering West Lafayette, Indiana 47907 Abstract—Three SIMD single stage interconnection networks which have been proposed and studied in the literature are the Illiac, PM2I, and Shuffle-Exchange. Here the ability of the Illiac and PM2I networks to per- form the shuffle interconnection in an SIMD machine with N processors is examined. A lower bound of 3\/N/2 transfers for the Illiac to shuffle data is derived. An algorithm to do this task in 2\/N-l transfers is given. A lower bound of log2N transfers for the PM2I to shuffle data has been published previously. An algo- rithm to do this task in log2N + l in transfers is presented here. 1. Introduction This paper extends SIMD interconnection network studies presented in [28, 31]. In particular, the ability of Fig. 1: PE-to-PE SIMD machine configuration, with the PM2I and Illiac single stage interconnection SIMD machine networks to perform the shuffle interconnection NPEs. is examined. In [28] it is shown that a lower bound on of configuration is shown in Fig. 1. It is called the PE- the number of transfers needed for the PM2I network to to-PE organization. The network is unidirectional and perform the shuffle is log2N, where N is the number of connects each PE to some subset of the other PEs. A processing elements in the SIMD machine. The algo- transfer instruction causes data to be moved from each rithm presented here requires only (log2N) + l transfers.
    [Show full text]
  • Simon E. Gluck Collection of Photographs of EDVAC and MSAC Computers 1990.232
    Simon E. Gluck collection of photographs of EDVAC and MSAC computers 1990.232 This finding aid was produced using ArchivesSpace on September 19, 2021. Description is written in: English. Describing Archives: A Content Standard Audiovisual Collections PO Box 3630 Wilmington, Delaware 19807 [email protected] URL: http://www.hagley.org/library Simon E. Gluck collection of photographs of EDVAC and MSAC computers 1990.232 Table of Contents Summary Information .................................................................................................................................... 3 Biographical Note .......................................................................................................................................... 3 Scope and Content ......................................................................................................................................... 4 Administrative Information ............................................................................................................................ 4 Related Materials ........................................................................................................................................... 5 Controlled Access Headings .......................................................................................................................... 5 Additonal Extent Statement ........................................................................................................................... 5 - Page 2 - Simon E. Gluck collection
    [Show full text]
  • P the Pioneers and Their Computers
    The Videotape Sources: The Pioneers and their Computers • Lectures at The Compp,uter Museum, Marlboro, MA, September 1979-1983 • Goal: Capture data at the source • The first 4: Atanasoff (ABC), Zuse, Hopper (IBM/Harvard), Grosch (IBM), Stibitz (BTL) • Flowers (Colossus) • ENIAC: Eckert, Mauchley, Burks • Wilkes (EDSAC … LEO), Edwards (Manchester), Wilkinson (NPL ACE), Huskey (SWAC), Rajchman (IAS), Forrester (MIT) What did it feel like then? • What were th e comput ers? • Why did their inventors build them? • What materials (technology) did they build from? • What were their speed and memory size specs? • How did they work? • How were they used or programmed? • What were they used for? • What did each contribute to future computing? • What were the by-products? and alumni/ae? The “classic” five boxes of a stored ppgrogram dig ital comp uter Memory M Central Input Output Control I O CC Central Arithmetic CA How was programming done before programming languages and O/Ss? • ENIAC was programmed by routing control pulse cables f ormi ng th e “ program count er” • Clippinger and von Neumann made “function codes” for the tables of ENIAC • Kilburn at Manchester ran the first 17 word program • Wilkes, Wheeler, and Gill wrote the first book on programmiidbBbbIiSiing, reprinted by Babbage Institute Series • Parallel versus Serial • Pre-programming languages and operating systems • Big idea: compatibility for program investment – EDSAC was transferred to Leo – The IAS Computers built at Universities Time Line of First Computers Year 1935 1940 1945 1950 1955 ••••• BTL ---------o o o o Zuse ----------------o Atanasoff ------------------o IBM ASCC,SSEC ------------o-----------o >CPC ENIAC ?--------------o EDVAC s------------------o UNIVAC I IAS --?s------------o Colossus -------?---?----o Manchester ?--------o ?>Ferranti EDSAC ?-----------o ?>Leo ACE ?--------------o ?>DEUCE Whirl wi nd SEAC & SWAC ENIAC Project Time Line & Descendants IBM 701, Philco S2000, ERA..
    [Show full text]
  • Online Sec 6.15.Indd
    6.155.9 Historical Perspective and Further Reading Th ere is a tremendous amount of history in multiprocessors; in this section we divide our discussion by both time period and architecture. We start with the SIMD approach and the Illiac IV. We then turn to a short discussion of some other early experimental multiprocessors and progress to a discussion of some of the great debates in parallel processing. Next we discuss the historical roots of the present multiprocessors and conclude by discussing recent advances. SIMD Computers: Attractive Idea, Many Attempts, No Lasting Successes Th e cost of a general multiprocessor is, however, very high and further design options were considered which would decrease the cost without seriously degrading the power or effi ciency of the system. Th e options consist of recentralizing one of the three major components. Centralizing the [control unit] gives rise to the basic organization of [an] . array processor such as the Illiac IV. Bouknight et al. [1972] Th e SIMD model was one of the earliest models of parallel computing, dating back to the fi rst large-scale multiprocessor, the Illiac IV. Th e key idea in that multiprocessor, as in more recent SIMD multiprocessors, is to have a single instruction that operates on many data items at once, using many functional units (see Figure 6.15.1). Although successful in pushing several technologies that proved useful in later projects, it failed as a computer. Costs escalated from the $8 million estimate in 1966 to $31 million by 1972, despite construction of only a quarter of the planned multiprocessor.
    [Show full text]
  • SIMD1 Ñ Illiac IV
    Illiac IV History Illiac IV n First massively parallel computer ● SIMD (duplicate the PE, not the CU) ● First large system with semiconductor- based primary memory n Three earlier designs (vacuum tubes and transistors) culminating in the Illiac IV design, all at the University of Illinois ● Logical organization similar to the Solomon (prototyped by Westinghouse) ● Sponsored by DARPA, built by various companies, assembled by Burroughs ● Plan was for 256 PEs, in 4 quadrants of 64 PEs, but only one quadrant was built ● Used at NASA Ames Research Center in mid-1970s 1 Fall 2001, Lecture SIMD1 2 Fall 2001, Lecture SIMD1 Illiac IV Architectural Overview Programming Issues n One CU (control unit), n Consider the following FORTRAN code: 64 64-bit PEs (processing elements), DO 10 I = 1, 64 each PE has a PEM (PE memory) 10 A(I) = B(I) + C(I) ● Put A(1), B(1), C(1) on PU 1, etc. n CU operates on scalars, PEs on vector- n Each PE loads RGA from base+1, aligned arrays adds base+2, stores into base, ● All PEs execute the instruction broadcast where “base” is base of data in PEM by the CU, if they are in active mode n Each PE does this simultaneously, giving a speedup of 64 ● Each PE can perform various arithmetic ● and logical instructions For less than 64 array elements, some processors will sit idle ● Each PE has a memory with 2048 64-bit ● words, accessed in less than 188 ns For more than 64 array elements, some processors might have to do more work ● PEs can operate on data in 64-bit, 32-bit, and 8-bit formats n For some algorithms, it may be desirable to turn off PEs n Data routed between PEs various ways ● 64 PEs compute, then one half passes data to other half, then 32 PEs compute, n I/O is handled by a separate Burroughs etc.
    [Show full text]
  • John William Mauchly
    John William Mauchly Born August 30, 1907, Cincinnati, Ohio; died January 8, 1980, Abington, Pa.; the New York Times obituary (Smolowe 1980) described Mauchly as a “co-inventor of the first electronic computer” but his accomplishments went far beyond that simple description. Education: physics, Johns Hopkins University, 1929; PhD, physics, Johns Hopkins University, 1932. Professional Experience: research assistant, Johns Hopkins University, 1932-1933; professor of physics, Ursinus College, 1933-1941; Moore School of Electrical Engineering, 1941-1946; member, Electronic Control Company, 1946-1948; president, Eckert-Mauchly Computer Company, 1948-1950; Remington-Rand, 1950-1955; director, Univac Applications Research, Sperry-Rand 1955-1959; Mauchly Associates, 1959-1980; Dynatrend Consulting Company, 1967-1980. Honors and Awards: president, ACM, 1948-1949; Howard N. Potts Medal, Franklin Institute, 1949; John Scott Award, 1961; Modern Pioneer Award, NAM, 1965; AMPS Harry Goode Memorial Award for Excellence, 1968; IEEE Emanual R. Piore Award, 1978; IEEE Computer Society Pioneer Award, 1980; member, Information Processing Hall of Fame, Infornart, Dallas, Texas, 1985. Mauchly was born in Cincinnati, Ohio, on August 30, 1907. He attended Johns Hopkins University initially as an engineering student but later transferred into physics. He received his PhD degree in physics in 1932 and the following year became a professor of physics at Ursinus College in Collegeville, Pennsylvania. At Ursinus he was well known for his excellent and dynamic teaching, and for his research in meteorology. Because his meteorological work required extensive calculations, he began to experiment with alternatives to mechanical tabulating equipment in an effort to reduce the time required to solve meteorological equations.
    [Show full text]
  • The Manchester University "Baby" Computer and Its Derivatives, 1948 – 1951”
    Expert Report on Proposed Milestone “The Manchester University "Baby" Computer and its Derivatives, 1948 – 1951” Thomas Haigh. University of Wisconsin—Milwaukee & Siegen University March 10, 2021 Version of citation being responded to: The Manchester University "Baby" Computer and its Derivatives, 1948 - 1951 At this site on 21 June 1948 the “Baby” became the first computer to execute a program stored in addressable read-write electronic memory. “Baby” validated the widely used Williams- Kilburn Tube random-access memories and led to the 1949 Manchester Mark I which pioneered index registers. In February 1951, Ferranti Ltd's commercial Mark I became the first electronic computer marketed as a standard product ever delivered to a customer. 1: Evaluation of Citation The final wording is the result of several rounds of back and forth exchange of proposed drafts with the proposers, mediated by Brian Berg. During this process the citation text became, from my viewpoint at least, far more precise and historically reliable. The current version identifies several distinct contributions made by three related machines: the 1948 “Baby” (known officially as the Small Scale Experimental Machine or SSEM), a minimal prototype computer which ran test programs to prove the viability of the Manchester Mark 1, a full‐scale computer completed in 1949 that was fully designed and approved only after the success of the “Baby” and in turn served as a prototype of the Ferranti Mark 1, a commercial refinement of the Manchester Mark 1 of which I believe 9 copies were sold. The 1951 date refers to the delivery of the first of these, to Manchester University as a replacement for its home‐built Mark 1.
    [Show full text]
  • Oral History Interview with David J. Wheeler
    An Interview with DAVID J. WHEELER OH 132 Conducted by William Aspray on 14 May 1987 Princeton, NJ Charles Babbage Institute The Center for the History of Information Processing University of Minnesota, Minneapolis Copyright, Charles Babbage Institute 1 David J. Wheeler Interview 14 May 1987 Abstract Wheeler, who was a research student at the University Mathematical Laboratory at Cambridge from 1948-51, begins with a discussion of the EDSAC project during his tenure. He compares the research orientation and the programming methods at Cambridge with those at the Institute for Advanced Study. He points out that, while the Cambridge group was motivated to process many smaller projects from the larger university community, the Institute was involved with a smaller number of larger projects. Wheeler mentions some of the projects that were run on the EDSAC, the user-oriented programming methods that developed at the laboratory, and the influence of the EDSAC model on the ILLIAC, the ORDVAC, and the IBM 701. He also discusses the weekly meetings held in conjunction with the National Physical Laboratory, the University of Birmingham, and the Telecommunications Research Establishment. These were attended by visitors from other British institutions as well as from the continent and the United States. Wheeler notes visits by Douglas Hartree (of Cavendish Laboratory), Nelson Blackman (of ONR), Peter Naur, Aad van Wijngarden, Arthur van der Poel, Friedrich L. Bauer, and Louis Couffignal. In the final part of the interview Wheeler discusses his visit to Illinois where he worked on the ILLIAC and taught from September 1951 to September 1953. 2 DAVID J.
    [Show full text]
  • Vector Machines  Vector Machines Today Introduction  a Vector Processor Is a CPU That Can Run One Instructiononanentire Vector of Data
    Hakam Zaidan Stephen Moore Outline Vector Architectures Properties Applications History Westinghouse Solomon ILLIAC IV CDC STAR 100 Cray‐1 Other Cray Vector Machines Vector Machines Today Introduction A Vector processor is a CPU that can run one instructiononanentire vector of data. The fetched number of instructions are small. They also achieve data parallelism in large scientific and multimedia applications. Styles of Vector Architectures Based on how the operands are fetched, vector processors can be divided into two categories: Memory‐Memory Architecture. Vector‐Register Architecture. Vector Processor Elements Vector Register: Fixed length, single vector, ports for reading and writing. Usually 8 to 32 registers of length 64 or 128 bits. Vector Functional Units (FUs): Usually 4‐8 functional units: FP mult, FP add, and FP divide, in addition to the integer add and logical shift Vector Load Store Unit (LSUs). Scalar registers. Cross‐bar. Vector Processor Properties Results are independent. Known pattern for memory access by the vector instructions. In pipelines, branches and branch problems are reduced. Single vector instruction indicate huge amount of calculations (e.g. loops). Disadvantages With scalar instructions: Relatively slow. Some difficulties in the implementation of the precise exceptions. High cost for on‐chip vector memory systems. Code complexity. Applications Lossy compression. Lossless compression. Multimedia Processing. Standard benchmarking kernels. Handwriting recognition. Speech recognition. Cryptography. Operating system and networking. Databases. Support of language run‐time. History In 1962, Illinois Automatic Computer series of super computers ILLIAC I, ILLIAC II, ILLIAC III, ILLIAC IV (with 64 ALUs 100‐ 150 Mflops). In 1973 TI’s Advance Scientific Computer (ASC) 20‐80 Mflops.
    [Show full text]
  • John W. Mauchly Papers Ms
    John W. Mauchly papers Ms. Coll. 925 Finding aid prepared by Holly Mengel. Last updated on April 27, 2020. University of Pennsylvania, Kislak Center for Special Collections, Rare Books and Manuscripts 2015 September 9 John W. Mauchly papers Table of Contents Summary Information....................................................................................................................................3 Biography/History..........................................................................................................................................4 Scope and Contents....................................................................................................................................... 6 Administrative Information........................................................................................................................... 7 Related Materials........................................................................................................................................... 8 Controlled Access Headings..........................................................................................................................8 Collection Inventory.................................................................................................................................... 10 Series I. Youth, education, and early career......................................................................................... 10 Series II. Moore School of Electrical Engineering, University of Pennsylvania.................................
    [Show full text]