duke compsci 220 / ece 252 advanced computer architecture i prof. alvin r. lebeck unit 0:...

Duke Compsci 220 / ECE 252Advanced Computer

Architecture I

Prof. Alvin R. Lebeck

Unit 0: Introduction

Compsci 220 / ECE 252 (Lebeck): Introduction 1

Slides developed by Amir Roth of University of Pennsylvania with sources that included University of Wisconsin slides by Mark Hill, Guri Sohi, Jim Smith, and David Wood.

Slides enhanced by Milo Martin, Mark Hill, Alvin Lebeck, Dan Sorin, and David Wood with sources that included Profs. Asanovic, Falsafi, Hoe, Lipasti, Shen, Smith, Sohi, Vijaykumar, and Wood

2

Instructors

• Prof: Alvin LebeckOffice: D308 LSRCHours: Tues 1:30 – 2:30, Wed 2:00-3:00 or by appointment (email)email: [email protected]

• Teaching Assistant (s)Vali PistolOffice: D339Hours: TBD or by appointmentemail: [email protected] IngallsHours: TBD or by appointmentEmail: [email protected]

Compsci 220 / ECE 252 (Lebeck): Introduction

3

Course Components

• Reading Materials Computer Architecture: A Quantitative Approach by Hennessy

and Patterson, 4th Edition

Recent research papers (on course website)

• Homework 5 or 6 homework assignments, performed in groups of 2

• Term Project Groups of 2 or 3

• Exams Midterm and final exam


4

Administrative (Grading)• 25% Homeworks (must be done in pairs w/

statement) 4 to 6 Homeworks Late < 1 day = 50% Late > 1 day = zero

• 50% Examinations (Midterm 20% + Final 30%)• 25% Research Project (work in groups of 3)

No late term projects

• Academic Misconduct University policy will be followed strictly Zero tolerance for cheating and/or plagiarism

• This course requires hard work.


5

Administrative (Continued)

• Midterm Exam: In class (75 min) closed book• Final Exam: (3 hours) closed book

• Big class, if you are going to drop, please do so early.


6

Administrative (Continued)

• Course Web Page http://www.cs.duke.edu/courses/cps220/fall10 Lectures posted there shortly before class (pdf) Homework posted there General information about course

• Blackboard Posting Grades only

• Course Forum Google groups – you will receive invite later this week Use it to

1. read announcements/comments on class or homework, 2. ask questions (help), 3. communicate with each other.


7

Reading

• H&P Chapter 1

• Papers Power a First-Class Architectural Design Constraint A Case for Energy-Proportional Computing

• After those:• Start: Appendix A (pipelining)


8

What you should Expect from Course

• Things NOT to expect in this course: 100% of class = me lecturing to you Homework sets and exams where every question is either

quantitative or has a single correct answer

• Things to expect in this course: Active discussions/arguments about architecture ideas Essay questions Being asked to explain, discuss, defend, and argue Questions with multiple possible answers


9

Expected Background

• Who’s in this class? PhD, MS, undergrad from Comp Sci, ECE, & other From all over the world…

• Courses you should have taken already Basic architecture (Compsci 104 / ECE 152 or equivalent) Programming (our simulator is in C) Basic OS (Compsci 110 / ECE 153) — not critical, but helpful

• Topics you should remember fondly - I will not cover these in detail in this course Instruction sets, computer arithmetic, assembly programming, I/O


10

Term Project

• This is a semester-long research project Do not expect to do whole thing in last week… I will likely define a default project, but many students choose to

pursue their own ideas Project proposals due Thursday, October 14

• You may “combine” this project with a project from another class, but you MUST consult with me first

• You must absolutely, positively reference prior work Do not copy text from any source (paper, book, internet, etc.) Please ask me if you have ANY questions Not knowing != valid excuse

Now on to computer architecture …


What is Computer Architecture?

• “Computer Architecture is the science and art of selecting and interconnecting hardware components to create computers that meet functional, performance and cost goals.” - WWW Computer Architecture Page

• An analogy to architecture of buildings…




Plans

The role of a building architect:

MaterialsSteel

ConcreteBrickWoodGlass Goals

FunctionCost

SafetyEase of Construction

Energy EfficiencyFast Build Time

Aesthetics

BuildingsHousesOffices

ApartmentsStadiumsMuseums

Design

Construction



Plans

The role of a computer architect:

“Technology”Logic Gates

SRAMDRAM

Circuit TechniquesPackaging

Magnetic StorageFlash Memory

GoalsFunction

PerformanceReliability

Cost/ManufacturabilityEnergy EfficiencyTime to Market

ComputerPCs

ServersPDAs

Mobile PhonesSupercomputersGame Consoles

Embedded

Design

Manufacturing

Three important differences: age (~60 years vs thousands), rate of change, automated mass production (magnifies design)

Design Goals• Functional

Needs to be correct What functions should it support (Turing completeness aside)

• Reliable Does it continue to perform correctly? Hard fault vs transient fault Google story - memory errors and sun spots Space probe vs PC reliability

• High performance “Fast” is only meaningful in the context of a set of important

tasks Not just “Gigahertz” Impossible goal: fastest possible design for all programs


Design Goals

• Low cost Per unit manufacturing cost (wafer cost) Cost of making first chip after design (mask cost) Design cost (huge design teams, why? Two reasons…)

• Low power Energy in (battery life, cost of electricity) Energy out (cooling and related costs) Static vs dynamic power, sleep modes, peak vs average Cyclic problem, very much a problem today

• Challenge: balancing the relative importance of these goals And the balance is constantly changing


Constant Change

• Absolute improvement, different rates of change, cyclic• Better computers help design the next generation (CAD)


“Technology”Logic Gates

SRAMDRAM

Circuit TechniquesPackaging

Magnetic StorageFlash Memory

Applications/DomainsPCs

ServersPDAs

Mobile PhonesSupercomputersGame Consoles

EmbeddedGoals

FunctionPerformance

ReliabilityCost/Manufacturability

Energy EfficiencyTime to Market

Rapid Change

• Exciting: perhaps the fastest moving field … ever Processors vs. cars

o 1985: processors = 1 MIPS, cars = 60 MPHo 2000: processors = 500 MIPS, cars = 30,000 MPH?

What if cars crashed as often as computers?

• Computation is driving most, if not all, fields Natural and physical sciences, social sciences, journalism, etc.


1971–1980

1981–1990

1991–2000

2008 2015

Transistors (M)

0.01–0.1 0.1–1 1–100 300-1000

10000?

Clock (MHz) 0.2–2 2–20 20–1000 3500 10000?

MIPS <0.2 0.2–20 20–2000 16000 100000?

First Microprocessor

• Connect a few transistors together to make…


• Intel 4004• 1971 (first

microprocessor)

• 4-bit data• 2300 transistors• 10 mm PMOS• 108 KHz• 12 V• 13 mm2

More Recent Microprocessor

• Or a few more to form…


• Intel Pentium4 + HT• 2003

• 32/64-bit data• 55M transistors• 0.90 mm CMOS• 3.4 GHz• 1.2 V• 101 mm2

By end of course, this will make sense!• Pentium 4 specifications: what do each of these

mean? Technology

o 55M transistors, 0.90 mm CMOS, 101 mm2, 3.4 GHz, 1.2 V Performance

o 1705 SPECint, 2200 SPECfp ISA

o X86+MMX/SSE/SSE2/SSE3 (X86 translated to RISC uops inside)

Memory hierarchyo 64KB 2-way insn trace cache, 16KB D$, 512KB–2MB L2o MESI-protocol coherence controller, processor consistency

Pipelineo 22-stages, dynamic scheduling/load speculation, MIPS

renamingo 1K-entry BTB, 8Kb hybrid direction predictor, 16-entry RASo 2-way hyper-threading


A recently announced processor

• Sun Rock 16 Microcores 1-2 threads per core 8 x 32KB L1D$ 4 x 32KB L1I$ 2MB L2$ 16MB L3$ 48 GB/s memory B/W 2.1 GHz 296 mm^2 321 million transistors 250W power Available 2H2009


Layers of abstraction

• Architects need to understand computers at many levels Instruction Set Architecture Microarchitecture Systems Architecture Technology Applications

• Good architects are “Jacks of most trades”


Instruction Set Architecture• Hardware/Software interface

Software impacto support OS functions

• restartable instructions• memory relocation and

protectiono a good compiler target

• simple• orthogonal

o Dense• Improve memory performance

Hardware impacto admits efficient implementation

• across generationso admits parallel implementation

• no ‘serial’ bottlenecksAbstraction without interpretation


OP R1 R2 R3 imm

OP R1M1 imm2R2M2

R3M3 imm2

Microarchitecture• Implement instruction set • Register-transfer-level (RTL)

design • Exploit capabilities of

technology locality and concurrency

• Iterative process generate proposed architecture estimate cost measure performance

• Still emphasis is on overcoming sequential nature of programs deep pipelining multiple issue dynamic scheduling branch prediction/speculation


Regs

Instr.Cache

IR

PC

BA

C

System-Level Design• Design at the level of

processors, memories, and interconnect.

• More important to application performance, cost and power than CPU (core) design

• Feeds and speeds constrained by IC pin count,

module pin count, and signaling rates

• System balance for a particular application

• Driven by performance/cost goals available components

(cost/perf) technology constraints


P

SW

400MHzDual Issue

16Bytes x 200MHz

I/O

M M M M

Disk

Net

Display

Large-system example

• Google Project 02: Oregon/Washington border• Cheap electricity: Columbia river• Cheap cooling: Columbia river• Cheap b/w: surplus optic network from tech-boom

era• Google, Yahoo, and Microsoft


Technology Trends• Processor (SRAM)

Density: ~30%, Speed: ~20%

• Memory (DRAM) Density: ~60%, Speed: ~4%

• Disk Density: ~60%, Speed: ~10%

• Changing quickly and with respect to each other!! Fundamentally changes design Different tradeoffs

• Exciting: constant re-evaluation and re-design


Shaping Force: Applications/Domains• Another shaping force: applications

Applications and application domains have different requirementso Domain: group with similar character

Lead to different designs

• Scientific: weather prediction, genome sequencing First computing application domain: naval ballistics firing tables Need: large memory, heavy-duty floating point Examples: CRAY T3E, IBM BlueGene

• Commercial: database/web serving, e-commerce Need: data movement, high memory + I/O bandwidth Examples: Sun Enterprise Server, AMD Opteron, Intel Xeon, IBM

Power 6


More Recent Applications/Domains• Desktop: home office, multimedia, games

Need: integer, memory bandwidth, integrated graphics/network? Examples: Intel Core 2, AMD Athlon

• Mobile: laptops Need: low power, integer performance, integrated wireless? Examples: Intel PentiumM

• Embedded: PDAs, cell phones, automobiles, door knobs Need: low power, low cost, integrated DSP? Examples: Intel StrongARM, X-Scale, Transmeta TM3200

• Sensors: disposable “smart dust” Need: extremely low power, extremely low cost


Application Specific Designs

• This course is mostly about general-purpose CPUs CPU that can do anything, specifically run a full OS E.g., Intel Core i3, i5, IBM Power6, AMD Opteron, Intel Itanium

• Large, profitable segment of application-specific CPUs Implement some critical domain-specific functionality in hardware

o Graphics engines, physics engines+ Much more effective (performance, power, cost) than software

+ General rule: hardware is better than software Examples: may or may not get to

o Emotion engine (PS2/PSX), Samsung phone, HDTV, iPod…


Why Study Computer Architecture?• Understand where computers are going

Future capabilities drive the computing world Forced to think 5+ years in the future

• Exposure to high-level design Less about “design” than “what to design” Engineering, science, art Architects paint with broad strokes The best architects understand all the levels

o Devices, circuits, architecture, compilers, applications• Understand hardware for software tuning• Real-world impact

no computer architecture no computers!• Get a job (design or research)


Some Course Goals• Exposure to the “big ideas” in computer architecture

• Exposure to examples of good (and some bad) engineering

• Understanding computer performance and metrics Empirical evaluation Understanding quantitative data and experiments

• “Research” exposure Read research literature (i.e., papers) Course project Cutting edge proposals

• Non-goals: “how computers work”, detailed design

• Let’s look at course home page…


duke compsci 220 / ece 252 advanced computer architecture i prof. alvin r. lebeck unit 0:...

Documents