Performance of Computers. EE365: Introduction to Digital Computer Design. So What? Ubiquity of Computers

Transcription

1 EE365: Introduction to Digital Computer Design Instructor: T.N. Vijaykumar T.A.: Tim Mattox Spring 1998 Purdue University Performance of Computers What do these two intervals have in common? (222 years) (2 years) Absolute speed improvements of computers comparable!! If performance improves by 50%, 1.5^2 = 2.25 Slide copies will be made available at HKN Lounge and Potter Library EE365 Lecture Notes 1 EE365 Lecture Notes: Chapter 1 2 Ubiquity of Computers So What? Originally used by the government to for military applications a few handful in number Now used by individuals for all kinds of things e.g., Norelco Shaver millions in number Higher performance leads to Better utilization of all those millions (and more) computers Newer applications considered infeasible previously EE365 Lecture Notes: Chapter 1 3 EE365 Lecture Notes: Chapter 1 4

2 Instructor Information T.N. Vijaykumar (Prof. Vijaykumar or simply vijay) 334B EE Building Introduction of EE365 EE266 - Transistors upto multiplexors EE362 - Assembly Language to Instruction Set Architecture EE365 - Puts the two together Requires managing complexity through ABSTRACTION EE365 Lecture Notes: Chapter 1 5 EE365 Lecture Notes: Chapter 1 6 Why Study Computer Design To design new computers: old designs obsolete fast new technologies - e.g., denser ICs new user demand - e.g., virtual reality To be an informed user a little auto mechanics helps owner rarely, but importantly Abstraction Difference between interface and implementation Interface - WHAT something does Implementation - HOW it does so To learn to deal with complexity via abstraction problems that take months and years to complete EE365 Lecture Notes: Chapter 1 7 EE365 Lecture Notes: Chapter 1 8

3 Abstraction - E.g., What s the Big Deal? 2-to-1 Mux Interface: X Y S Mux O S O 0 X 1 Y E.g., HyperSPARC interface book Worse for computers, in general - a tower of abstraction Application software System software (OS and compiler/assembler/linker) Hardware (CPU, memory, I/O) Each interface is complex and implemented with layer below Abstraction keeps unnecessary details hidden Implementations gates (fast or slow), pass transistors Hundreds of engineers to build one product EE365 Lecture Notes: Chapter 1 9 EE365 Lecture Notes: Chapter 1 10 In space and time Basic Division of Hardware In time Basic Division of Hardware Control Data path Processor In space Memory Output Input Fetch the instruction from memory add r1, r2, r3 Decode the instruction - what does this mean? Read input operands read r2, r3 Perform operation add Write results write to r1 Determine next instruction pc := pc + 4 EE365 Lecture Notes: Chapter 1 11 EE365 Lecture Notes: Chapter 1 12

4 Why don t old designs work? Technology changes in non-uniform ways Date What Comments st transistor Bell Labs st Integrated Circuit Texas Instruments st microprocessor Intel 1974 Intel transistors 1978 Intel K transistors 1989 Intel M transistors 1995 Intel Pentium Pro 5.5M transistors Why don t old designs work Application needs change Missile trajectories Payroll processing Spread sheats Desktop publishing Virtual reality Most likely, you can dream up something better EE365 Lecture Notes: Chapter 1 13 EE365 Lecture Notes: Chapter 1 14 Classes of Computers The Future? Supercomputer $5 million - 20 million Mainframe $1 million - 4 million Minicomputer $20 thousand thousand PC/Workstation $2 thousand - 20 thousand Network Computer/WebTV $ Embedded computer $1-10 ( invisible like electric motor) Transistors 6 M 350 M Clock 200 MHz 4 GHz Performance What can we do with this power? Natural language What can we do embedded? Wrist watch assistant YOU will be making it happen EE365 Lecture Notes: Chapter 1 15 EE365 Lecture Notes: Chapter 1 16

5 Classroom change Class has been moved to PHYS 203 Typo in approximate outline Building computer chips Complex multi-step process slice ingots -> wafers process wafers (many steps) -> patterned wafers dice patterned wafers -> dies test dies -> good dies bond good die to package -> packaged dies (parts) test parts -> good parts ship to customers -> make money! EE365 Lecture Notes: Chapter 1 17 EE365 Lecture Notes: Chapter 1 18 Performance vs. Design Time Time to deliver product is important E.g., a new design will take 3 years to complete will be 3 times faster but if technology improves 50% per year in 3 years 1.5^3 = 3.38 so new design is worse! Bottomline Designers must know BOTH software and hardware Both contribute to layers of abstraction of computers IC costs and performance Compilers and Operating Systems EE365 Lecture Notes: Chapter 1 19 EE365 Lecture Notes: Chapter 1 20