When and Where? Course Information. Expected Background ECE 486/586. Computer Architecture. Lecture # 1. Spring Portland State University

Transcription

1 When and Where? ECE 486/586 Computer Architecture Lecture # 1 Spring 2015 Portland State University When: Tuesdays and Thursdays 7:00-8:50 PM Where: Willow Creek Center (WCC) 312 Office hours: Tuesday and Thursday after class, or by appointment TA: Yafei Yang (yafeiat pdxdot edu) Webpage: Go to the course webpage for: Class slides Course syllabus Course schedule Homework assignments and solutions Course Information Expected Background Textbook:Computer Architecture: A Quantitative Approach, 5 th Edition, John L.Hennessy anddavid A. Patterson. Morgan Kauffman, ISBN Supplemental materials (papers, articles, tutorials) for certain topics will be provided by the instructor ECE485/585 or equivalent Basic knowledge of computer organization (processor, memory, I/O) Design of a simple uniprocessor Basic knowledge of instruction set architecture Programming experience in a high-level language (for example, C or C++) needed for course project We ll also cover research papers and articles published in computer architecture conference/journals Homework assignments may require paper reading If you lack some of this background, please refer to Computer Organization and Design: The Hardware/Software Interface, 4 th Edition, Hennessy and Patterson

2 Grading Policy Homeworks 20% 4 homework assignments, each worth 5% Class Project 20% Midterm Exam 25% Final Exam 35% Grading scale (tentative): A: % A-: % B+: % B: % B-: % C+: % C: % C-: % D+: % D: % D-: % F: Below 50% Other Policies All homework assignments due in class. No extensions Electronic homework submission (via to instructor) allowed Submission must be received before the class start time Midterm exam in class during week 6 Tuesday, May 5, 7:00 8:30 PM Final exam will cover entire course with more emphasis on material taught after the midterm exam Final project will require simulation of architecture features in software. More details will be provided later Interactive Lectures Interaction reinforces the learning process Please raise your hand and ask if something is not clear No question is stupid or trivial No confusion is small or unimportant Don t think everybody else gets it except you If you didn t get it then probably some others didn t either By asking for clarification, you are helping everyone! What this course is all about? Computer Architecture = Instruction Set Architecture + Computer Organization + Computer Hardware This course builds upon itself, so if something is not clear please ask right away or you will not understand later lectures

3 Computer Architecture Definitions Instruction Set Architecture (ISA): Set of instructions visible to the programmer, boundary between hardware and software Computer Organization (or Microarchitecture): High-level aspects and structure of the computer Computer Hardware:Detailed logic design, packaging, process technology Computer Architecture Definitions (cont.) Historically, the term computer architecture referred to ISA, Currently, hardware and organization matter more than ISA Examples: Same ISA, different organization AMD Opteron and Intel Core i7 Both implement x86 ISA but have very different pipelines and cache organizations Same ISA, different hardware Intel Core i7 and Intel Xeon 7560 Both implement x86 ISA and have nearly identical organizations But offer different clock rates and different memory systems Why study Computer Architecture? Why study Computer Architecture? Answer # 1:Technology advancements require constant re-optimization of cost/power/performance Technology Annual Improvement Transistor Count 25% Transistor Speed 20-25% DRAM Density 60% DRAM Speed 4% Disk Density 25% Disk Speed 4% Answer # 2: Innovation needed to continue existing trends Moore s law Original version: Transistor scaling exponential Popular version: Processor performance doubling every 18 months Initially transistor counts limited performance ~ 35% performance improvement per year Now, larger transistor counts enable advanced microarchitectures ~ 50% growth per year Added growth due to implementation/organization

4 Why study Computer Architecture? Answer # 3:Changing user requirements demand new technologies and architectures Example: Portability Longer batter life Energy-efficient architectures Previously infeasible solutions become high-volume products! Portable computing Touch-based interfaces Virtual reality Data mining Gesture recognition Course Topics Technology and Economics of Computer Design Measuring, reporting and summarizing computer performance Instruction set architectures Processor datapath and pipelining Instruction-level parallelism Branch Prediction Advanced topics in processor architecture History of Processor Performance Performance Contributors Process technology + architectural enhancements 17 years of sustained performance growth at a rate of > 50% per year ~ 35% per year from process technology and circuits, the rest from computer architecture and microarchitecture Effects of sustained performance growth: More compute capability available to individual users New classes of computers (PCs, tablets, smartphones, warehouse computers) Mainframes/minicomputers replaced by microprocessor-based computers Programmers focus on productivity rather than on performance

5 Microprocessor Segments Personal Mobile Devices Personal Mobile Devices (PMD) Smartphones Tablets Desktop and Laptop Computing PCs (Windows, Linux) Workstations (higher performance) E.g., running CAD applications Servers Clusters/Warehouse-scale Computers Embedded Computers Price of system: $ $1000 Largest computing segment in dollar terms Price of microprocessor: $10 -- $100 Cost Performance Responsiveness Battery life, no fans Desktops and Laptops Servers Price of system: $ $2500 Price of microprocessor: $50 -- $500 More than half of this segment is battery-operated laptops Performance per dollar Graphics performance: critical for gaming markets Well-characterized applications and benchmarks Increasing use of internet-centric applications poses new challenges for performance evaluations Price of system: $ $10,000,000 Price of microprocessor: $ $2,000 WWW and e-commerce have accelerated growth Traditional back office applications E-commerce: Amazon.com, Ebay, E*trade etc. Throughput (transactions per minute or web pages served per second) Availability (e.g., consider the servers running bank ATM machines) Scalability (e.g., consider the rise in number of Facebook users)

6 Clusters/Warehouse-scale Computers Price of system: $100, $200,000,000 Price of microprocessor: $50 -- $250 Clusters are collections of desktops/servers connected by LANs, acting as a single larger computer Different from servers in that clusters use redundant, inexpensive components Performance per dollar Throughput Proportionality (80% of the maintenance cost is associated with power and cooling) Embedded Computers Price of system: $10 -- $100,000 Price of microprocessor: $ $100 Largest segment by unit volume Ubiquitous Automobiles, appliances, printers, set-top boxes, industrial controls, networking switches Price Real-time performance Application-specific performance Relative Size of Markets Task of the Computer Architect Successful computer architect must design a computer to meet functional requirementsas well as price, power, performanceand availability goals Determine what features are needed for a market segment Incorporate architecture features (instructions, microarchitecture optimizations) that would make the computer competitive in that market Example: SSE instructions in x86 architectures for DSP and graphics processing Identify important technology trends and adapt computer design accordingly Example:Error-correction codes in caches to mitigate manufacturing imperfections Source: IDC

7 Processor Performance: Future Outlook Since 2003, annual processor performance growth limited to < 22%: Power dissipation constraints of air-cooled chips limit on clock speed Lack of instruction-level parallelism in a uniprocessor Market drivers dictate power (energy) and cost to be as important as (or more than) performance Future performance improvements would come from multiple processors per chip rather than via faster uniprocessors Parallelism is now the driving force of computer design across all market segments, with energy and cost being the primary constraints Parallel Architectures: Classes of Parallelism Instruction-level Parallelism Multiple instructions from the same thread executed in parallel by using techniques like pipelining and speculative execution Thread-level Parallelism Multiple threads of work are created and can operate independently and largely in parallel Data-level Parallelism Same instruction(s) operated on many data items at the same time (SIMD, MIMD)