Advanced Computer Architecture - Lecture 2: Quantitative principles

Advanced Computer Architecture - Lecture 2: Quantitative principles. This lecture will cover the following: detailed discussion on the computer performance – the key to quantitative design and analysis; growth in processor performance; price-performance design; CPU performance metrics;...

CS 704 Advanced Computer Architecture

Lecture 2

Quantitative Principles

Detailed discussion on the computer Performance – the key to

Summary

Recap of Lecture 1

Computer Systems:

Architecture refers to those attributes of a

computer visible to a programmer or compiler

writer; e.g instruction set, addressing techniques, I/O mechanisms etc

Organization refers to how the features of a

computer are implemented? i.e., control signals

Recap of Lecture 1

Computer Development:

•Academically , modern computer developments have

their infancy in 1944-49

•Commercially, the first machine was built by

Eckert-Mauchly Computer Corporation in 1949

•Technological developments , from vacuum tubes to VLSI circuits, dynamic memory and network technology gave birth to four different generations of computers

•Microprocessor and PCs were introduced in 1971

Recap of Lecture 1

Design Perspectives:

Processor – ISA, ILP and Cache Memory hierarchy: Multilevel

cache and Virtual memory

input/output and storages

MAC/VU-Advanced Computer Architecture 6

Recap of Lecture 1

Computer Design Cycle:

• The computer design and development has been under the influence of

Growth in Processor Performance

Insert Slide 9 here

•The supercomputers and mainframes , costing

millions of dollars and occupying excessively

large space, prevailing form of computing in

1960s were replaced with relatively low-cost and

smaller-sized minicomputers in 1970s

•In 1980s , very low-cost microprocessor-based

desktop computing machines in the form of

Growth in Processor Performance

Insert Slide 9 here

•The growth in processor performance since

mid-1980s has been substantially high than in earlier years

•Prior to the mid-1980s microprocessor

performance growth was averaged about 35% per year

•By 2001 the growth raised to about 1.58 per

year

Growth in Processor Performance

Alpha

MIPS R2000

DEC Alpha

Price-Performance Design

Technology improvements are used to lower the cost and increase performance The relationship between cost and

price is complex one

The cost is the total amount spends to

produce a product

The price is the amount for which a

finished good is sold

Price-Performance Design

The cost passes through

different stages before it

becomes price

A small change in cost may

have a big impact on price

Price vs Cost … Insert Slide 14 here

• Manufacturing Costs: Total amount spent to

produce a component

- Component Cost: Cost at which the

components are available to the designer - It ranges from 40% to 50% of the list price of the product

- Recurring costs: Labor, purchasing

scrap, warranty – 4% - 16 % of list price

- Gross margin – Non-recurring cost: R&D,

marketing, sales, equipment, rental,

maintenance, financing cost, pre-tax profits, taxes

Price vs Cost … Insert Slide 14

here

• List Price :

• Amount for which the finished good is

sold;

• it includes Average Discount of

15% to 35% of the as volume discounts and/or retailer markup

Price vs Cost … Price-Performance Design Cont’d

Cost-effective IC Design: Price-Performance Design

components surviving testing

Volume: increases manufacturing hence decreases the list price and improves the purchasing efficiency

transistor or wire in either x or y direction

Cost-effective IC Design: Price-Performance Design

Reduction in feature size from 10 microns in

1971 and 0.18 in 2001has resulted in:

- Quadratic rise in transistor count

- Linear increase in performance

- 4-bit to 64-bit microprocessor

- Desktops have replaced time-sharing

machines

Cost of Integrated Circuits

The Integrated circuit manufacturing passes through many stage:

Wafer chopping it into dies

Cost of Integrated Circuits

Insert Slide 19 here Die: is the square area of the wafer containing the integrated circuit

See that while fitting dies on the wafer the

waist

Cost of a die: The cost of a die is determined

from cost of a wafer; the number of dies fit

on a wafer and the percentage of dies that work, i.e., the yield of the die

Dies of Integrated Circuits

Cost of Integrated Circuits

Insert Slide 21 here

• The cost of integrated circuit can be determined as ratio of the total cost ;

i.e., the sum of the costs of die, cost of testing die, cost of packaging and the cost of final testing a chip; to the final test yield

Calculating Integrated Circuits Costs

die cost + die testing cost + packaging cost + final testing cost

final test yield

Cost of Integrated Circuits

Insert Slide 23 here

per wafer and die yield

Calculating Integrated Circuits Costs

die cost + die testing cost + packaging cost + final testing cost

final test yield

dies per wafer x die yield

Cost of Integrated Circuits

Insert Slide 25 here

Calculating Integrated Circuits Costs

die cost + die testing cost + packaging cost + final testing cost

final test yield

dies per wafer x die yield

π (wafer diameter/2)2 π (wafer diameter)

Example Calculating Number of Dies

For die of 0.7 Cm on a side, find the number of dies per wafer

Calculating Die Yield Insert Slide 29 here

• Die yield is the fraction or percentage of

good dies on a wafer number

• Wafer yield accounts for completely bad

wafers so need not be tested

• Wafer yield corresponds to on defect

density by α which depends on number of masking levels

• good estimate for CMOS is 4.0 and

Calculating Integrated Circuits Costs

Price-Performance Design

• Time to run the task:

• Execution time, response time, latency

• Throughput or bandwidth:

• Tasks per day, hour, week, sec, ns …

Price-Performance Design

Insert Slid 32

• Example:

• To carry 2400 passengers from Lahore to Islamabad –

• Train completes the task in 4:00 hrs while airplane completes the same task

in 6.00 hrs.;

• e., 66.67% of the task in same time – throughput and hence performance of train is 50% more than airplane

Price-Performance Design: Example

Vehicle

Train

Plane

Cost / person

300 Rs

3000 Rs

Time Lah to Isb

4.0 hours

45 min

Passenge rs/ trip

2400

300

Execution time /person

6.0 sec

9.0 sec

Cost-performance

300x6=1,800Rs-sec/person

3000x9=27,000Rs-sec/person

Time to complete job

4.0 hours

45x8 min

= 6.0 Hr

Plane 10 time faster but takes

50% more time to complete the

job; i.e., lesser throughput –

thus performance of train is

50%better than plane

The time per person and cost person of train is less than that of plane Thus the cost-performance of plane

is 1:15

Metrics of Performance

Insert Slide 33

Megabytes per second

Compiler

Programming Language

Application

Instruction Set Architecture

Answers per month Operations per second

Datapath Control Function

Aspects of CPU Performance

Inst Count CPI Clock Rate

Program √

Compiler √

Inst Set √ √

Organization √ √

Technology √

Cycles Per Instruction

= CPU Clock Cycles for program / Instruction Count

= (CPU Time * Clock Rate) / Instruction Count

For instruction mix, the relative frequency of occurrence of different types of instructions is given as:

FICi = IC of ith instruction / Total Instruction count

Example: Calculating average CPI

Base Machine (Reg / Reg)

Cycles Per Instruction

Geometric mean time:

n

/ n

/ π    Execution time ratio

Summary: Price-Performance Design

Computer cost:

The total cost of manufacturing a computer is distributed among different parts of the system such as the cost of cabinet,

processor board and I/O devices

Performance Time is the key measurement of performance

Comparing performance of two designs: the ratio ,

η = Execution time Y / Execution time X

determines how much lower execution time machine Y takes as

compared to X ; as performance is inverse of execution time, i.e.,

η = Performance X / Performance Y

Instruction Execution Rate - MIPS

MIPS specify performance inversely to execution time;

For a given program:

MIPS = (instruction count) / (execution time x 106)

MIPS could not be calculated from the instruction mix

reference machine as:

or

= [Time reference / Time M] x MIPS reference

CPU Benchmark Suites

Performance Comparison: the execution time of

the same workload running on two machines without

running the actual programs

Benchmarks: the programs specifically chosen to measure the performance

Five levels of programs: in the decreasing order

SPEC: System Performance Evaluation Cooperative

First Round 1989: 10 programs yielding a single number – SPECmarks

Second Round 1992: SPECInt92 (6 integer programs) and SPECfp92 (14 floating point programs)

Third Round 1995

– new set of programs: SPECint95 (8 integer programs) and

SPECfp95 (10 floating point)

– “benchmarks useful for 3 years”

Summary: Designing and performance comparison

• Designing to Last through Trends

Capacity Speed

• Time to run the task

– Execution time, response time, latency

• Tasks per day, hour, week, sec, ns, …

– Throughput, bandwidth

• “X is n times faster than Y” means

ExTime(Y) Performance(X)

= ExTime(X) Performance(Y)

Summary …… Cont’d

CPI Law:

Execution time is the REAL measure of computer performance!

Good products created when have:

– Good benchmarks, good ways to summarize

performance

Die Cost goes roughly with die area4

Summary … Cont’d

“For better or worse, benchmarks shape a field”

Good products created when have:

– Good benchmarks

– Good ways to summarize performance

Given sales is a function in part of performance relative to competition, investment in improving product as reported

by performance summary

If benchmarks/summary inadequate, then choose between improving product for real programs vs improving product

to get more sales;

Sales almost always wins!

Execution time is the measure of computer performance!