Lecture 1: Course Overview and Intro to High Performance Processors

Date: Sept. 27

Background Material

• Matrix Multiplication
• C Programming

Reading

• Chapters 1-3 of High Performace Computing.

Topics

• Course Overview and Objectives
• A simple benchmark
  • Review Matrix multiplication and block matrix multiplication
  • Timing program
  • Timing results (using different compiler options for optimization)
  • Reporting results (processor information)
• Overview of High Performance Processors
  • RISC vs. CISC
  • Pipelining
  • Hazards and Delays
  • Delayed branches and branch prediction
  • ISA influence on pipelining
  • Superscalar execution and superpipelines
  • Out of order and speculative computation
• Overview of High Performance Memory Design
  • Memory/processor gap
  • Memory technology
  • Memory hierarchy
  • Cache - principle of locality
  • Cache organization
    1. Direct-mapped
    2. Fully associative
    3. set-associative
    4. line size
    5. Write policy
  • Virtual memory
    1. Mapping and page tables
    2. TLB
    3. Page faults
  • Improving memory preformance
    1. Increased size and width
    2. Multiple caches
    3. Cache bypass
    4. Interleaving
    5. Prefetching
    6. Benefits of speculative computation
  • Dynamic RAM technology trends
    1. Extended data out (EDO) RAM
    2. Synchronous DRAM (SDRAM)
    3. RAMBUS
    4. Cached DRAM (CDRAM)

  Lecture notes

  • Lecture Slides (lec1.ppt)

  Programs

  • Matrix benchmark
    1. matrix.h
    2. matrix.c
    3. block.c
    4. timeblockmm.c

  Related Links

  • CS281 (System Architecture I)
  • ATLAS (Automatically Tuned Linear Algebra Software)
  • PHIPAC (Portable High Performance ANSI C)
  • Netlib Performance Database (look for linpack benchmark)

  Assignments

  • Assignment 1 (Due in class on Oct. 4 - next class)
  Created: Sept. 27, 2000 by jjohnson@mcs.drexel.edu