1. Lecture 2: Intro to Computer Architecture
Michael B. Greenwald
Computer Architecture
CIS 501
Fall 1999

2. General Information
Class: TR 1:30-3, in LRSM Auditorium Recitation: T 10: in Moore 225
Instructor: Professor Michael Greenwald Office: Moore (GRW), room Office hours: R10:30-12noon or by appt.
TA: Sotiris Ioannidis Office: Moore, room 102e Office hours: TR5-6PM or by appt.
Secretary: Christine Metz Office: Moore, room 556

3. Outline Review Quantitative principles of computer design
Amdahl’s law
CPU performance equation
Quantitative measurements
Costs
Performance

4. Typos in HW 3c. New version on web page. D = defects/
Defects per layer

5. Technology Trends: Microprocessor Capacity
“Graduation Window”
Alpha 21264: 15 million
Pentium Pro: 5.5 million
PowerPC 620: 6.9 million
Alpha 21164: 9.3 million
Sparc Ultra: 5.2 million
Moore’s Law
CMOS improvements:
Die size: 2X every 3 yrs
Line width: halve / 7 yrs

6. Trends in application demands
Program increase memory demands by factor of per year (1/2 to 1 bit/year)
Avail. disk space (or net bw) is always consumed.
User I/O bandwidth grows: tty->crt->bitmap->video->?virtual reality?
Processing power: cheapest to produce one version of program. Optimize for mid-range. Slow on low- end, fast on high-end.
Are these demands growing because of increased capabilities or increased appetites?

7. The Quantitative Approach

8. Measurement and Evaluation Quantitative Approach
Architecture is an iterative process:
Searching the space of possible designs
At all levels of computer systems
Cost /
Performance
Analysis
Creativity
Good Ideas
Mediocre Ideas
Bad Ideas

9. Measurement and Evaluation Quantitative Approach
Not a guarantee of good ideas, just a way to discard bad ideas.
Cost /
Performance
Analysis
Creativity
Good Ideas
Mediocre Ideas
Bad Ideas

10. Computer Engineering Methodology
Technology
Trends

11. Computer Engineering Methodology
Evaluate Existing
Systems for
Bottlenecks
Benchmarks
Where to start: existing systems bottlenecks
Technology
Trends

12. Computer Engineering Methodology
Evaluate Existing
Systems for
Bottlenecks
Benchmarks
Technology
Trends
Simulate New
Designs and
Organizations
Workloads

13. Computer Engineering Methodology
Evaluate Existing
Systems for
Bottlenecks
Implementation
Complexity
Benchmarks
How hard to build
Importance of simplicity (wearing a seat belt); avoiding a personal disaster
Theory vs. practice
Technology
Trends
Implement Next
Generation System
Simulate New
Designs and
Organizations
Workloads

14. Measurement Tools Benchmarks, Traces, Mixes
Hardware: Cost, delay, area, power estimation
Simulation (many levels)
ISA, RT, Gate, Circuit
Queuing Theory
Rules of Thumb
Fundamental “Laws”/Principles
Measure
Experiment
Analyze
Design

15. All produce “measures”: what do measures mean? How do they compare?
Measurement Tools
Benchmarks, Traces, Mixes
Hardware: Cost, delay, area, power estimation
Simulation (many levels)
ISA, RT, Gate, Circuit
Queuing Theory
Rules of Thumb
Fundamental “Laws”/Principles
Measure
Experiment
Analyze
Design
All produce “measures”: what do measures mean? How do they compare?

16. The Bottom Line: Performance (and Cost)
Plane
DC to Paris
6.5 hours
3 hours
Speed
610 mph
1350 mph
Passengers
470
132
Throughput (pmph)
286,700
178,200
Boeing 747
Fastest for 1 person?
Which takes less time to transport 470 passengers?
BAD/Sud Concorde
Time to run the task (ExTime)
Execution time, response time, latency
Tasks per day, hour, week, sec, ns … (Performance)
Throughput, bandwidth

17. The Bottom Line: Performance (and Cost)
Plane
DC to Paris
6.5 hours
3 hours
Speed
610 mph
1350 mph
Passengers
470
132
Throughput (pmph)
286,700
178,200
Boeing 747
Fastest for 1 person?
Which takes less time to transport 470 passengers?
BAD/Sud Concorde
Which is better?

18. The Bottom Line: Performance (and Cost)
Plane
DC to Paris
6.5 hours
3 hours
Speed
610 mph
1350 mph
Passengers
470
132
Throughput (pmph)
286,700
178,200
Boeing 747
Fastest for 1 person?
Which takes less time to transport 470 passengers?
BAD/Sud Concorde
Which is better? It depends if you are trying to win a race from DC to Paris, or you are trying to move the most people.

19. The Bottom Line: Performance (and Cost)
Plane
DC to Paris
6.5 hours
3 hours
Speed
610 mph
1350 mph
Passengers
470
132
Throughput (pmph)
286,700
178,200
Boeing 747
Fastest for 1 person?
Which takes less time to transport 470 passengers?
BAD/Sud Concorde
Even if trying to move most people, performance is useless without understanding cost. Else, why not just fly two Concordes at once, doubling throughput?
, $160M in ‘98

20. Costs
Performance metrics are mostly useless without understanding costs.

21. Integrated Circuits Costs
IC cost = Die cost Testing cost Packaging cost
Final test yield
Die cost = Wafer cost
Dies per Wafer * Die yield
Wafer
Defect
Die
Smaller dies are cheaper, and reduce cost per defect.

22. Integrated Circuits Costs
IC cost = Die cost Testing cost Packaging cost
Final test yield
Die cost = Wafer cost
Dies per Wafer * Die yield
Defect
Smaller dies are cheaper, and reduce cost per defect.

23. Die Cost goes roughly with die area4
IC Cost parameters
 Number of masking levels (measure of manufacturing complexity), was typically 3.0, growing
wafer yield = wafers that are not completely bad. Typically close to 100%
Defects per unit area = 0.6 to 1.2 per cm2. Drops with learning curve.
Die Cost goes roughly with die area4

24. Integrated Circuits Costs
IC cost = Die cost Testing cost Packaging cost
Final test yield
Die cost = Wafer cost
Dies per Wafer * Die yield
Dies per wafer =  * ( Wafer_diam / 2)2 –  * Wafer_diam – Test dies
Die Area  2 * Die Area
Die Yield = Wafer yield * 1 +

Defects_per_unit_area * Die_Area  { }
Die Cost goes roughly with die area4

25. Integrated Circuits Costs
Die cost = Wafer cost
Dies per Wafer * Die yield
Dies per wafer =  * ( Wafer_diam / 2)2 –  * Wafer_diam – Test dies
Die Area  2 * Die Area
Die Yield = Wafer yield * 1 +
Die Cost = Wafer cost * 1 +
 * ( Wafer_diam / 2)2 –  * Wafer_diam

Defects_per_unit_area * Die_Area  { } 
Defects_per_unit_area * Die_Area  { }
Die Cost goes roughly with die area4

26. Die Cost goes roughly with die area+1
IC Cost parameters
Defects per unit area = 0.6 to 1.2 per cm2
Technologies that can fix defects (e.g. lasers a’la Lincoln Labs (MIT)), reduce effective defects per unit area and increase yield. However, need to understand costs which differ from formula.
Still:
Die Cost goes roughly with die area+1

27. Real World Examples(circa ‘93)
Chip Metal Line Wafer Defect Area Dies/ Yield Die Cost layers width cost /cm2 mm2 wafer
386DX $ % $4
486DX $ % $12
PowerPC $ % $53
HP PA $ % $73
DEC Alpha $ % $149
SuperSPARC $ % $272
Pentium $ % $417
From "Estimating IC Manufacturing Costs,” by Linley Gwennap, Microprocessor Report, August 2, 1993, p. 15

28. Other Costs Die Test Cost = Test Jig Cost * Ave. Test Time Die Yield
Packaging Cost: depends on pins, heat dissipation
Chip Die Package Test & Total cost pins type cost Assembly
386DX $ QFP $1 $4 $9
486DX2 $ PGA $11 $12 $35
PowerPC 601 $ QFP $3 $21 $77
HP PA $ PGA $35 $16 $124
DEC Alpha $ PGA $30 $23 $202
SuperSPARC $ PGA $20 $34 $326
Pentium $ PGA $19 $37 $473

29. Cost/Performance What is Relationship of Cost to Price?
Component Costs
Direct Costs (add 25% to 40%) recurring costs: labor, purchasing, scrap, warranty
Gross Margin (add 82% to 186%) nonrecurring costs: R&D, marketing, sales, equipment maintenance, rental, financing cost, pretax profits, taxes
Average Discount to get List Price (add 33% to 66%): volume discounts and/or retailer markup
List Price
Average
Discount
25% to 40%
Avg. Selling Price
Gross
Margin
34% to 39%
6% to 8%
Direct Cost
Component
Cost
15% to 33%

30. Cost/Performance What is Relationship of Cost to Price?
Component Costs
Direct Costs (add 25% to 40%) recurring costs: labor, purchasing, scrap, warranty
Gross Margin (add 82% to 186%) nonrecurring costs: R&D, marketing, sales, equipment maintenance, rental, financing cost, pretax profits, taxes
Average Discount to get List Price (add 33% to 66%): volume discounts and/or retailer markup
List Price
Average
Discount
Avg. Selling Price
Discretion
Gross
Margin
Direct Cost
Component
Cost

31. Chip Prices (August 1993) Assume purchase 10,000 units
Chip Area Mfg. Price Multi- Comment
mm2 cost plier
386DX 43 $9 $ Intense Competition
486DX2 81 $35 $ No Competition
PowerPC $77 $
DEC Alpha 234 $202 $ Recoup R&D?
Pentium 296 $473 $ Early in shipments

32. Summary: Price vs. Cost

33. Cost/Price/Profit How is R&D funded?
R&D 4% to 12%, contributes to gross margin (it is an indirect cost)
Two views:
Only 4% of income on R&D!
Investment: every $1 spent on R&D should lead to $8 to $25 in sales!

34. PERFORMANCE

35. The Bottom Line: Performance (and Cost)
Plane
DC to Paris
6.5 hours
3 hours
Speed
610 mph
1350 mph
Passengers
470
132
Throughput (pmph)
286,700
178,200
Boeing 747
Fastest for 1 person?
Which takes less time to transport 470 passengers?
BAD/Sud Concorde
Even if trying to move most people, performance is useless without understanding cost. Else, why not just fly two Concordes at once, doubling throughput?
, $160M in ‘98

36. Performance Terminology
Time versus Performance: duration vs. rate.
Time: response time = execution time
Rate: throughput
Reciprocals: there is both a time and a performance measure for any performance metric.
“Improve performance”: time decreases, performance increases
For computer systems the key performance metric is total execution time

37. Meaning of “Execution Time” (a.k.a. Response time)
Wall-clock-time, response time, elapsed-time: latency (including idle time)
vs. CPU Time: non-idle
System vs. User time: both elapsed and CPU
system performance: elapsed time on unloaded system (includes OS + idle time)
CPU performance: user CPU time on unloaded system

38. Terminology
What do we mean when we compare two measures and say that “X is n times faster than Y”?

39. The Bottom Line: Performance (and Cost)
"X is n times faster than Y" means
ExTime(Y) Performance(X)
= = n
ExTime(X) Performance(Y)
Speed of Boeing 747 vs. Concorde
Throughput of Boeing 747 vs. Concorde
1350 / 610 = 2.2X
286,700/ 178, X

40. The Bottom Line: Performance (and Cost)
"X is n times faster than Y" means
286,700 Performance(X)
= 1.60
178,200 Performance(Y)
Speed of Boeing 747 vs. Concorde
Throughput of Boeing 747 vs. Concorde
1350 / 610 = 2.2X
286,700/ 178, X

41. The Bottom Line: Performance (and Cost)
"X is n times faster than Y" means
286,700 Performance(X)
= 1.60
178,200 Performance(Y)
Speed of Boeing 747 vs. Concorde
Throughput of Boeing 747 vs. Concorde
1350 / 610 = 2.2X
286,700/ 178, X
Note: Natural or meaningful units. Hours per passenger-mile is slightly weirder than passenger-miles per hour.

42. Measurement Tools Benchmarks, Traces, Mixes
Hardware: Cost, delay, area, power estimation
Simulation (many levels)
ISA, RT, Gate, Circuit
Queuing Theory
Rules of Thumb
Fundamental “Laws”/Principles
Measure
Experiment
Analyze
Design
ENGINEERING:Convert this to that

43. Fundamental Principle of Computer Design
Make the common case fast
In every trade-off, favor the frequent case over the infrequent case.
But how do we quantify this? At what point is the cost to the infrequent case sufficiently large as to offset speedups to the frequent case?

44. Fundamental Principle of Computer Design
Make the common case fast
In every trade-off, favor the frequent case over the infrequent case.
But how do we quantify this? At what point is the cost to the infrequent case sufficiently large as to offset speedups to the frequent case?
Amdahl’s Law quantifies this principle

45. Amdahl's Law Speedup due to enhancement E:
ExTime w/o E Performance w/ E
Speedup(E) = =
ExTime w/ E Performance w/o E
Suppose that enhancement E accelerates a fraction F of the task by a factor S, and the remainder of the task is unaffected

46. Amdahl's Law Speedup due to enhancement E:
ExTime w/o E Performance w/ E
Speedup(E) = =
ExTime w/ E Performance w/o E
Suppose that enhancement E accelerates a fraction F of the task by a factor S, and the remainder of the task is unaffected

47. Amdahl’s Law
ExTimenew = ExTimeold x (1 - Fractionenhanced) + Fractionenhanced
Speedupenhanced 1
ExTimeold
ExTimenew
Speedupoverall = =
(1 - Fractionenhanced) + Fractionenhanced
Speedupenhanced

48. Amdahl’s Law: Example
Floating point instructions improved to run 2X; but only 10% of actual instructions are FP
ExTimenew =
Speedupoverall =

49. Amdahl’s Law: Example
Floating point instructions improved to run 2X; but only 10% of actual instructions are FP
ExTimenew = ExTimeold x ( /2) = 0.95 x ExTimeold 1
Speedupoverall = =
1.053
0.95

50. Amdahl’s Law: Example
Suppose fetching a page from a web cache is 1000 times faster than getting the page over the net, but hit rate on cache is only 30%
ExTimenew =
Speedupoverall =

51. Amdahl’s Law: Example
Suppose fetching a page from a web cache is 1000 times faster than getting the page over the net, but hit rate on cache is only 30%
ExTimeWCache =
ExTimeold x ( /1000) = ExTimeold x .7003 1
.7003
Speedupoverall = =
1.428

52. Amdahl’s Law: Example
Just because something seems quantifiable, doesn’t mean it is meaningful.
Quality of class = .5 student effort quality of instructor value of material
MetricWProf+ =
Metricold x ( /106)  Metricold x .75 1
.75
Speedup = =
1.333
So even if I were a million times better as a professor, the class would only be times as good.