Page 1 1 CS6810 School of Computing University of Utah Computer Architecture CS/ECE 6810 Today’s topics: • course logistics & motivation • computer architecture as a profession • market segments • technology scaling and cost 2 CS6810 School of Computing University of Utah Introductory Material • A few tidbits on the instructor • Pay close attention to the course web page http://www.eng.utah.edu/~cs6810 it will change so keep up to date on what’s there • Why study computer architecture? it’s required for CS grad students » OK I get that but will try to make it interesting anyway for SW types: » understanding the architecture maximize code performance for HW types: » essential understanding for the profession rich area where contributions are badly needed » one of which might be your thesis current state of the art is in a wild time » architecture changes directions – see “badly needed” » lots of job opportunitiesPage 2 3 CS6810 School of Computing University of Utah Artifacts of Rapid Change • Textbook 4th edition is significantly more relevant than previous versions » BUT it’s now 2-3 years old and a lot has happened » result: lectures will have some disagreements w/ the text • reading the book will be necessary • attending the lectures will hopefully also be valuable the basic issues in the text are still important » bulk of course will focus on this material tons of research literature » not a requirement in this course but helpful to clarify or deepen your understanding » the internet is your friend • as is the University’s subscription to digital libraries – IEEE Xplore is probably the most useful – ACM is a good 2nd choice 4 CS6810 School of Computing University of Utah Computer Architecture • Strictly speaking – it’s a whole system thing study of the structure of computer hardware » requires a diverse set of systems & circuit understanding • languages & operating systems • high level organizational issues (our focus in CS6810) – processor, cache, main memory, I/O, networking/interconnect, storage • analysis via tools such as simulation – power, performance, energy efficiency, verification • transistor circuits, wires, and fabrication technology • layout, EDA tools, cooling, packaging, … » you can’t be a wizard in one of these areas • without understanding the constraints and interfaces imposed by the other disciplines • The profession: industry: design & build the systems of the future » often w/ large teams of specialized wizards academic: study and explore new directions » few actually build things except as models via simulationPage 3 5 CS6810 School of Computing University of Utah A Snippet of Modern History • Mechanical difference engine proposed in 1786 by J. H. Mueller 2 versions built by Charles Babbage in the 1820’s » image at right is a replica in the computer museum in Mtn. View, CA • Electronic computer WW2: army needed something to compute ballistics tables contract w/U Penn in 1943 operational in 1946 » analog machine » programmed by plugging cables into the right spot » YOW!! numerous analog machines follow » vacuum tubes, crystal diodes, … 6 CS6810 School of Computing University of Utah History II • Stored program computer Concept: EDVAC report 1945 by John von Neumann » hence the von Neumann architecture tag Implementation » Jun 1948 – Manchester Univ. experimental machine • “Baby” SSEM (small scale experimental machine) • memory = Williams Cathode Ray Tubes » First practical: EDSAC May 1949 (also done in the UK) • memory = delay-linesPage 4 7 CS6810 School of Computing University of Utah Then Came Transistors … • First all transistor computer MIT’s Linc, TX-0, TX-2 – Wes Clark 1950’s » led to networks, graphics, interactive computing • Integrated circuit more than one transistor on a die – 1958 » Rob’t Noyce and Gordon Moore @ Fairchild • later founders of Intel first microprocessor Intel 4004 in 1971 » 10 um pMOS, 92 Kips, 740 KHz, 4-bit data-path, BCD » it’s been a wild ride ever since 8 CS6810 School of Computing University of Utah Unprecedented Improvement • Moore’s surprising prediction in 1965 holds up reasonably well so farPage 5 9 CS6810 School of Computing University of Utah CAGR Inflection Points • Improvement consistent technology gain architecture less consistent • Inflection points 1st 25 years: 25% due to both late 70’s uProcessors emerge » 35% » + 17% from architecture • RISC, pipelining, ILP and multiple issue (a.k.a. super scalar) » 16 years of Moore’s law growth 2002 things slow to ~20% » 3 key hurdles: thermals, insufficient ILP, slow memory • DRAM improvement trends: CAGR = 7% – latency hiding worked well until 2002 New agenda: TLP and DLP » enter multi-threading and multi-core architectures 10 CS6810 School of Computing University of Utah Computer Classes/Market Segments • Note today all classes are microprocessor based » not all microprocessors are the same • even when they appear to be the same to the programmer also my classes are quite different than the text’s • Embedded (fastest growth segment) huge range: automotive, cell phones, … large internet switches specialization » CISCO EPS-1 already contains 192 core » processors vary: • 4- 64-bit processors – price from a few cents to a few hundred dollars – system cost from $1 to $1M » typical differentiation • typical user tends to not be the programmer – provides a relatively fixed function or service • hard or soft real time performance often requiredPage 6 11 CS6810 School of Computing University of Utah Segments (cont’d) • Netbook cheap, light, and a bigger screen than a cell phone » battery life is a key issue • processor performance compromised for energy efficiency • Laptop a bit heavier and more expensive » more diversity in performance and energy efficiency than
View Full Document