CSU, Chico: Computer Science 380
(Computer Architecture)
Enlarge
|
Normal Size
|
Reduce
Slide
of 68 (Refresh off)
Ch 1: FUNDAMENTALS OF COMPUTER DESIGN
Microprocessor performance trends
Current computing markets
What is Computer Architecture?
According to H&P (2003) ...
Technology trends
The Thinning of the Wires
Primary goal of Computer Architecture?
A general engineering principle
Marketability in Computer Architecture
Decreasing cost trends (DRAM)
Ex: Pentium III cost trends
Ex: Intel Pentium 4 microprocessor die
Ex: 8-inch MIPS64 R20K wafer
Ex: Cost vs. price for a $1K PC
"For historical reasons ..."
More "legacy" examples
Limits on traditional models
Example: Pentium 4 die (color)
Example: Pentium 4 chip area breakdown
Cost and yield
Yield and fault-tolerance
Performance metrics
Example: Code size <> performance
Measures of time
Benchmarks
SPEC Benchmark Suite: CPU2000
SPEC Benchmark Suite: HPC2002, OMP
SPEC Benchmark Suite: viewperf, apc
SPEC Benchmark Suite: sfs, web99, web99ssl
TPC Benchmarks: TPC-C, TPC-H
TPC Benchmarks: TPC-R, TPC-W
The EEMBC Benchmarks
SysOpt.com - System optimization information
Reporting performancs = reproducibility
Comparing benchmark results
Averaging: arith mean vs weighted arith mean
Normalized execution time
Geometric mean for normalizing
Weighting example (arith vs geom means)
General systems engineering (GSE): variables
GSE: total unit cost
GSE: per-use cost
GSE: generalized cost component
GSE: additive cost expressions
Generalized Amdahl's Law
Law of Diminishing Returns
Practical implications
Focus on execution time
Focus of Amdahl's Law proper
Amdahl's Law - H&P Forms
Graphical visualization
Review: CPU performance equations
Review: cache memory systems
Review: computing total stall time
Typical memory hierarchy (mid 1990s)
Performance and performance per unit price
Performance and price-performance, int benchmark
Performance and price-performance, FP benchmark
High-performance servers
Good price-performance servers
Embedded processor performance
Embedded processor price-performance
Power consumption & efficiency as a metric
Performance per watt
P4 performance relative to P3
Peak versus observed performance
Fallacies and Pitfalls
This
E-Slideshow
was prepared by
Dr. J
for
CSCI 380
(Last revised: Thu Jan 30 21:34:39 PDT 2003)