ECE 462562 Computer Architecture and Design TTh 12

ECE 462/562 Computer Architecture and Design T-Th 12: 30 -1: 45 in HARV 210 www. ece. arizona. edu/~ece 462 Instructor Name: Ali Akoglu (ece. arizona. edu/~akoglu) Office: ECE 356 -B Phone: (520) 626 -5149 Email: akoglu@ece. arizona. edu Office Hours: Tuesdays 11: 00 AM – 12: 00 PM Thursdays 11: 00 AM- 12: 00 AM or by appointment

Computer Architecture Algorithm Programming Language Operating System/Virtual Machines Instruction Set Architecture (ISA) Gates/Register-Transfer Level (RTL) Circuits Devices Physics Abstraction Layers Application 2

Computer Architecture is Design and Analysis Architecture is an iterative process: • Searching the space of possible designs • At all levels of computer systems Creativity Cost / Performance Analysis Good Ideas Bad Ideas Mediocre Ideas 3

Computer Architecture Applications suggest how to improve technology, provide revenue to fund development Applications Technology Co mp atib ility Improved technologies make new applications possible Cost of software development makes compatibility a major force in market 4

Trends: The End of the Uniprocessor Era P! IL , re a tw f so nd a re Ha ba sed Ha are rdw a dw r Intel cancelled high performance uniprocessor, joined IBM and Sun for multiple processors 5

Crossroads: Conventional Wisdom • Old Conventional Wisdom: Power is free, Transistors expensive • New Conventional Wisdom: “Power wall” Power expensive, Xtors free (Can put more on chip than can afford to turn on) • Old CW: Sufficiently increasing Instruction Level Parallelism via compilers, innovation (Out-of-order, speculation, VLIW, …) • New CW: “ILP wall” law of diminishing returns on more HW for ILP • Old CW: Multiplies are slow, Memory access is fast • New CW: “Memory wall” Memory slow, multiplies fast (200 clock cycles to DRAM memory, 4 clocks for multiply) • Old CW: Uniprocessor performance 2 X / 1. 5 yrs • New CW: Power Wall + ILP Wall + Memory Wall = Brick Wall • • Uniprocessor performance now 2 X / 5(? ) yrs Sea change in chip design: multiple “cores” (2 X processors per chip / ~ 2 years) • More simpler processors are more power efficient 6

Instruction Set Architecture: Critical Interface software instruction set hardware • Properties of a good abstraction – Lasts through many generations (portability) – Used in many different ways (generality) – Provides convenient functionality to higher levels – Permits an efficient implementation at lower levels 7

ISA vs. Computer Architecture • Old definition of computer architecture = instruction set design • Other aspects of computer design called implementation • Our view is computer architecture >> ISA • Architect’s job much more than instruction set design; technical hurdles today more challenging than those in instruction set design • What really matters is the functioning of the complete system – hardware, runtime system, compiler, operating system, and application • Computer architecture is not just about transistors, individual instructions, or particular implementations 8

Course Focus Understanding the design techniques, machine structures, technology factors, evaluation methods that will determine the form of computers in 21 st Century Technology Applications Operating Systems Parallelism Programming Languages Computer Architecture: • Organization • Hardware/Software Boundary Measurement & Evaluation Interface Design (ISA) Compilers History 9

Related Courses ECE 568 Parallel Processing ECE 369 Strong Prerequisite Basic computer organization, first look at pipelines + caches ECE 462/562 Computer Architecture, First look at parallel architectures ECE 569 High Performance Computing, Advanced Topics ECE 474/574 ECE 576 Computer Aided Logic Design, FPGAs Computer Based Systems 10

Introduction • Text for ECE 462/562: Hennessy and Patterson’s Computer Architecture, A Quantitative Approach, 5 th Edition • Topics 1. Simple machine design (ISAs, microprogramming, unpipelined machines, Iron Law, simple pipelines) 2. Memory hierarchy (DRAM, caches, optimizations) plus virtual memory systems, exceptions, interrupts 3. Complex pipelining (score-boarding, out-of-order issue) 4. Explicitly parallel processors (vector machines, VLIW machines, multithreaded machines) 5. Multiprocessor architectures (memory models, cache coherence, synchronization) 11

Your ECE 462/562 • How would you like your ECE 462/562? • Mix of lecture vs. discussion • Depends on how well reading is done before class • Goal is to learn how to do good systems research • Learn a lot from looking at good work in the past • At commit point, you may chose to pursue your own new idea instead. 12

Coping with ECE 462/562 • • • Undergrads must have taken ECE 274 and ECE 369 Grad students with too varied background • Review Appendix A, B, C review of ISA, Datapath, Pipelining and Memory Hierarchy 13

Policies • Background: ECE 369 or equivalent, based on Patterson and Hennessy’s Computer Organization and Design • Prerequisite: ECE 274 & ECE 369 & Programming in C • 3 to 4 assignments, 2 exams, final project • Grad students: extra exam questions, survey paper and presentation • NO LATE ASSIGNMENTS • Make-ups may be arranged prior to the scheduled activity. • Inquiries about graded material => within 3 days of receiving a grade. • You are encouraged to discuss the assignment specifications with your instructor, and your fellow students. However, anything you submit for grading must be unique and should NOT be a duplicate of another source. • Read before the class • Participate and ask questions • Manage your time • Start working on assignments early 14

Grading Distribution of Components Grades Scale Component Percentage Grade Assignments+Quiz +Participation 35 90 -100% A Exam-I 15 80 -89% B Exam-II 15 70 -79% C Project 35 60 -69% D Total 100 Below 60% E

Introduction • Assignments and Project Ø Pairs only Ø Who is my partner? (email by 09/06) • Assignment-0 due 08/28 • Announcements on the web 16

Research Paper Reading • • • As graduate students, you are now researchers Most information of importance to you will be in research papers Ability to rapidly scan and understand research papers is key to your success 17

Project (Undergrad vs Grad) • • Transition from undergrad to grad student ECE wants you to succeed, but you need to show initiative pick topic (more on this later) meet 3 times with faculty to see progress give oral presentation (grad students only) written report like conference paper 3 weeks work full time for 2 people Opportunity to do “research in the small” to help make transition from good student to research colleague 18

Project (Undergrad vs Grad) • • Recreate results from research paper to see • If they are reproducible • If they still hold • Papers from ISCA, HPCA, MICRO, IPDPS, ISC Performance evaluation of an architecture • Using industry sponsored tools • GEM 5: gem 5. org • Pin: pintool. org • Simple. Scalar: simplescalar. com A complete end-to-end processor (UGs !!) • Take advantage of FPGAs!! Propose your own research project that is related to computer architecture 19

Measuring Performance • Topics: (Chapter 1) Ø Technology trends ØPerformance equations 20

Technology Trends and This Book 1996 When I took this class! 2002 2009 2011 Reduced ILP to 1 chapter! Shift to multicore! Reduced emphasis on ILP Request, Data, Thread, Introduce thread level P. Instruction Level Introduce: GPU, cloud computing, Smart phones, 21 tablets!

Problems • • • Algorithms, Programming Languages, Compilers, Operating Systems, Architectures, Libraries, … not ready to supply Thread Level Parallelism or Data Level Parallelism for 1000 CPUs / chip, Architectures not ready for 1000 CPUs / chip • Unlike Instruction Level Parallelism, cannot be solved by just by computer architects and compiler writers alone, but also cannot be solved without participation of computer architects 5 th Edition Computer Architecture: A Quantitative Approach explores shift from Instruction Level Parallelism to Thread Level Parallelism / Data Level Parallelism 22

Classes of Parallelism • • In Applications Ø Data Level Parallelism o Data items that can be operated on concurrently Ø Task-level Parallelism o Tasks of a work can operate independently In Hardware Ø ILP: exploits DLP with compiler, pipelining, speculative execution Ø Vector Architectures and GPUs: exploit DLP by applying a single instruction to a collection of data Ø Thread-level parallelism: exploits DLP and TLP, tightly coupled hardware, interaction among threads Ø Request level parallelism: exploits largely decoupled tasks specified by the programmer 23

Processor Technology Trends • Shrinking of transistor sizes: 250 nm (1997) 130 nm (2002) 65 nm (2007) 32 nm (2010) 28 nm(2011, AMD GPU, Xilinx FPGA) 22 nm(2011, Intel Ivy Bridge, die shrink of the Sandy Bridge architecture) • Transistor density increases by 35% per year and die size increases by 10 -20% per year… more cores! 24

Trends: Historical Perspective 25

Power Consumption Trends • Dyn power a activity x capacitance x voltage 2 x frequency • Capacitance per transistor and voltage are decreasing, but number of transistors is increasing at a faster rate; hence clock frequency must be kept steady • Leakage power is also rising • Power consumption is already between 100 -150 W in high-performance processors today Ø 3. 3 GHz Intel core i 7: 130 watts 26

Recent Microprocessor Trends Transistors: 1. 43 x / year Cores: 1. 2 - 1. 4 x Performance: 1. 15 x Frequency: 1. 05 x Power: 1. 04 x 2004 2010 Source: Micron University Symp. 27

Improving Energy Efficiency Despite Flat Clock Rate • Turn off the clock of inactive modules • Disable FP unit, core, etc. • Dynamic Voltage-Frequency Scaling • Periods of low activity, lower the clock rate • Low power mode • DRAMs lower power mode for extending the battery • Overclocking • Intel, Turbo mode (2008), chip decides safe clock rate • i 7 3. 3 GHz, can run in short bursts for 3. 6 GHz 28

Modern Processor Today • Intel Core i 7 q Clock frequency: 3. 2 – 3. 33 GHz q 45 nm and 32 nm products q Cores: 4 – 6 q Power: 95 – 130 W q Two threads per core q 3 -level cache, 12 MB L 3 cache q Price: $300 - $1000 29

Other Technology Trends • DRAM density increases by 40 -60% per year, latency has reduced by 33% in 10 years, bandwidth improves twice as fast as latency decreases • Disk density improves by 100% every year, latency improvement similar to DRAM 30

First Microprocessor Intel 4004, 1971 • 4 -bit accumulator architecture • 8 mm p. MOS • 2, 300 transistors • 3 x 4 mm 2 • 750 k. Hz clock • 8 -16 cycles/inst. 31

Hardware • Team from IBM building PC prototypes in 1979 • Motorola 68000 chosen initially, but 68000 was late • 8088 is 8 -bit bus version of 8086 => allows cheaper system • Estimated sales of 250, 000 • 100, 000 s sold [ Personal Computing Ad, 11/81] 32

DYSEAC, first mobile computer! • Carried in two tractor trailers, 12 tons + 8 tons • Built for US Army Signal Corps 33

Measuring Performance • Two primary metrics: wall clock time (response time for a program) and throughput (jobs performed in unit time) • To optimize throughput, must ensure that there is minimal waste of resources • Performance is measured with benchmark suites: a collection of programs that are likely relevant to the user § SPEC CPU 2006: cpu-oriented programs (for desktops) § SPECweb, TPC: throughput-oriented (for servers) § EEMBC: for embedded processors/workloads 34

Performance CPU time = Seconds = Instructions x Cycles x Seconds Program Instruction Cycle Program Inst Count X CPI Compiler X X Inst. Set. X X Organization X Technology Clock Rate X X 35

Amdahl’s Law • Architecture design is very bottleneck-driven – make the common case fast, do not waste resources on a component that has little impact on overall performance/power • Amdahl’s Law: performance improvements through an enhancement is limited by the fraction of time the enhancement comes into play 36

Amdahl’s Law • Considering an enhancement that runs 10 times faster than the original machine but is only usable 40% of the time. • Only 1. 56 x overall speedup • An application is “almost all” parallel: 90%. Speedup using • 10 processors => 5. 3 x • 100 processors => 9. 1 x • 1000 processors => 9. 9 x 37

Principle of Locality • Most programs are predictable in terms of instructions executed and data accessed • Temporal locality: a program will shortly re-visit X • Spatial locality: a program will shortly visit X+1 38

Exploit Parallelism • Most operations do not depend on each other – hence, execute them in parallel • At the circuit level, simultaneously access multiple ways of a set-associative cache • At the organization level, execute multiple instructions at the same time • At the system level, execute a different program while one is waiting on I/O 39