CSCI-564 Advanced Computer Architecture

Transcription

1 CSCI-564 Advanced Computer Architecture Lecture 1: Introduction Bo Wu Colorado School of Mines

2 Disclaimer: most of the slides in this course are adapted from four top-notch computer architecture researchers: Onur Mutlu (CMU) Steven Swanson (UCSD) Rajeev Balasubramonian (Utah) David Wentzlaff (Princeton)

3 Today s Agenda o What is architecture? o Why is it important? o What s in this class? o A bit history of computing 3

4 What is architecture? o How do you build a machine that computes? Quickly, safely, cheaply, efficiently, etc. o Architects develop new mechanism for performing and organizing mechanical computation 4

5 What is architecture? o How do you build a machine that computes? Quickly, safely, cheaply, efficiently, etc. o Architects develop new mechanism for performing and organizing mechanical computation 4

6 Why is architecture important? o For the world Computer architecture provides the engines that power all of computing 5

7 Why is architecture important? o For the world Computer architecture provides the engines that power all of computing 5

8 Why is architecture important? o For the world Computer architecture provides the engines that power all of computing o For you As computer scientists, software engineers, and sophisticated users, understanding how computers work is essential The processor is the most important piece of this story Many performance problems have their roots in architecture 6

9 Different scales 7

10 A server 8

11 MacBook Air 9

12 iphone 4S 10

13

14 From sand to applications Cool things happen here 12

15 Abstractions of the physical world 13

16 for the rest of the system 14

17 Current state of architecture

18 Moore s Law o # of transistors we can build in a fixed area of silicon doubles every two years 16

19 Intel 4004

20 Since

21 Evidence 19

22 The end of clock speed scaling o Clock speed is the biggest contributor to power Doubling the clock speed increases power by 2-8X Clock speed scaling is essentially finished o Most future performance improvements will be due to architectural and process technology improvements Indicates that computer architecture research is more important than ever 20

23 Power and heat 21

24 The rise of parallelism o Multi-processors If one CPU is fast, two must be faster! They allow you to (in theory) double performance without changing the clock speed. o Seems simple, so why are becoming so important now Speeding up a single CPU makes everything faster! An application s performance double every 18 months with no effort on the programmer s part Getting performance out of multiprocessors requires work Parallelizing code is difficult, it takes ( lots of ) work 22

25

26 AMD Barcelona

27 Power consumption trends o Dynamic power is proportional to activity x capacitance x voltage 2 x frequency o Capacitance per transistor and voltage are decreasing o Leakage power is rising o Energy = power x time = (dynpower + lkgpower) x time o Question (true or false): low power design is always good? 25

28 Power saving techniques o Dynamic voltage scaling Good: voltage is a big contributor to power Bad: it hurts performance Application scenarios? o Shut down the power supply Commonly used by cellphones Good: obvious Bad: May hurt performance significantly 26

29 A research problem Imagine yourself as a server of 10 cores. Every 0.1s you receive a request, which can be served by one core using 0.1s. You can optimize for power using the techniques we discussed. What kind of things you can do to reduce power consumption?

30 Goals for this class o Understand the trends shaping architecture today 28

31 Logistics

32 Course Staff o Instructor: Bo Wu (bwu@mines.edu) Lectures Tues + Thurs Office hours Wed. 3:30PM-5PM or by appointment o TA: Kai Liu (kaliu@mines.edu) o Course website: CSCI_564_15SPRING/info.shtml 30

33 More about me o I work on high performance computing and compiler optimization. o I am recruiting graduate students to work with. Talk to me if you are interested. o For teaching, I promise you I would not do this: 31

34 More about me o I work on high performance computing and compiler optimization. o I am recruiting graduate students to work with. Talk to me if you are interested. o For teaching, I promise you I would not do this: 31

35 Academic honesty o Do not cheat Do not copy other people s code Do not copy solutions during the exams Do not copy sentences from other papers or web resources Seriously, cheating leads to unhappy consequences 32

36 What you need to do o Class participation 10% Not showing up will impact your grade Read the text! I randomly pick students to ask basics about the reading o Homework 20% Four assignments o Two exams: Midterm 15% Final 15% 33

37 Projects 34

38 Projects o For PhD students Two projects Project 1: prefetching competition (two rounds) 20% Project 2: open-ended research project 20% 34

39 Projects o For PhD students Two projects Project 1: prefetching competition (two rounds) 20% Project 2: open-ended research project 20% o For all other students Option 1: Project: prefetching competition (two rounds) 40% Option 2: Do what is required for PhD students and get 2 bonus points for your final grade 34

40 Research projects o Why? Doing research is fun You are graduate students, who are supposed to do some o How? Choose a problem with my help Read a lot of related papers Work on my/your own ideas and produce some results 35

41 Sample research projects from CSCI-580 (Advanced HPC) Large-scale Graph Processing on GPUs (Matt, Zach and John) state-of-the-art: 4.739s our approach: 1.76s 36

42 Sample research projects from CSCI-580 (Advanced HPC) Parallel Execution of Finite State Machines on GPUs (Austin and Aruna)

43 Sample research projects from CSCI-580 (Advanced HPC) Parallelization of a Super Optimizer Implementation (Troy and Brianna)

44 An incomplete history of computation

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60