Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

CKV

Advanced VLSI Architecture


MEL G624

Lecture 3: Quantitative Analysis


CKV

Performance
CPU clock cycles Processors executes Instructions

¿ Instructions for a program × Average clock cycles per instruction

Clock cycles per instruction  CPI

Classic CPU performance equation


CPU time
¿ No of Instructions × CPI ×Clock cycle time

No of instructions × CPI
¿
Clock Rate
CKV

Performance

Alternative Representation

CPU time
¿ No of Instructions × CPI ×Clock cycle time

No of instructions × CPI
¿
Clock Rate

Instructions Clock cycles Seconds


time=seconds/ program=¿ program × Instruction ×Clock Cycle
CKV

Trends in Cost

Cost sensitive designs are of growing significance

Understanding of cost and its factors essential for


designers

Manufacturing process critical to cost

The underlying principle that drives the cost down is


the learning curve
CKV

Trends in Cost

Process
CKV

Trends in Cost

Cost of Integrated circuit

cost of die+cost of testing die +cost of packaging∧final test


Cost of IC =
Final test yield

Focus on cost of die

Cost of wafer
Cost of die=
dies per wafer × Die yield
CKV

Trends in Cost

wafer area
dies per wafer =
Die area

OR

2
wafer diameter

dies per wafer =


(
π×
2 )
Die area

Is this accurate ? What about dies near the periphery ?


CKV

Trends in Cost

Accurately
2
wafer diameter

dies per wafer =


(
π×
2 ) −( π × wafer diameter )
Die area √2 × Die area

This gives the max number of dies per wafer.

How many of these dies are good ? Die Yield


CKV

Trends in Cost

Bose-Einstein formula
Wafer yield
Die yield = N
(1+ Defects per unit area × Die area )
Empirical model [Sydow 2006]

Wafer yield – helps in eliminating the wafers that are completely bad

Defects per unit area – measure of random manufacturing defects


0.016-0.057 defects per square cm for 40 nm in 2010

N – parameter called process-complexity factor,


11.5-15.5 for 40 nm in 2010
CKV

Trends in Cost

Examples for a 300 mm wafer


109 good 2.25 cm2 dies
424 good 1.00 cm2 dies

Processing 300 mm wafer cost $3000-$6000 ( 3-4 Lakhs INR)

Assuming cost of $5500


Cost 1.00 cm2 die would be about $13

Cost 2.25 cm2 die would be about $51

Doubling the area Quadruples the cost


CKV

Trends in Cost

Other significant costs


Testing and packaging
Mask cost for low volumes

Guess the cost of mask (6-7 metal layers) ?

Cost of operation- servers


CKV

Quantitative principles of Computer Design

So far
Trends in Technology

Performance

Cost

What are the guidelines and principles that are useful in


computer design ?
CKV

Quantitative principles of Computer Design


Guidelines for improved design
Take advantage of parallelism
System level parallelism – Multicore
Instruction level parallelism – Pipelining
Hardware level parallelism – e.g. Carry look ahead vs ripple

Principle of Locality
Programs tend to reuse instructions and data – cache, virtual
Mem
Focus on Common case
In making design tradeoff favor frequent case over
infrequent case
CKV

Quantitative principles of Computer Design


Improving one aspect of computer will increase overall
performance by same amount

Do you think its correct ?

Suppose a program runs in 100 seconds on a computer with


multiply operations responsible for 80 seconds of the time. How
much do I have to improve the speed of multiplication if I want my
program to run 5 times faster ?
CKV

Quantitative principles of Computer Design


Amdahl’s law
E xecution time after improvement=¿
Execution time affected by improvement
+ Execution time unaffected
Amount of imrovement

E xecution time before improvement =Executionold  

E xecution time after improvement= Executionnew

Fraction of the old execution time that is affected by imrovement =Fractionenhanced


Amount of I mrovement =speedup enhanced

E xecution time after improvement= Executionnew =¿ Complete the equation


CKV

Quantitative principles of Computer Design


Amdahl’s law

Fraction enhanced
Execution new=Executionol d ×( ( 1 − Fraction enhanced ) + )
speedupenhanced

E xecution time before improvement =Executionold  

E xecution time after improvement= Executionnew

F raction of the old execution time that is affected by imrovement =Fractionenhanced


Amount of I mrovement =speedup enhanced
CKV

Quantitative principles of Computer Design

Fraction enhanced
Execution new=Executionol d ×( ( 1 − Fraction enhanced ) + )
speedupenhanced

Execution old 1
Speedup overall= =
Execution new Fraction enhanced
( 1 − Fraction enhanced ) +
speedup enhanced

Overall speed up is limited by


CKV

What is Computer Architecture ?

Evolution of computers

Power and performance

Cost

Amdahl’s Law
CKV

Problems
A compiler designer is trying to decide between two code sequences for a
particular computer. The hardware designer have supplied the following
facts. CPI for each instruction class
A B C
CPI 1 2 3

For a particular high level language statement, the compiler writer is


considering two code sequences that require the following instruction
counts Code Instruction Counts
Sequence
A B C

1 2 1 2
2 4 1 1
Which code sequence executes the most instructions ? Which will be
faster ? What is the CPI for each sequence ?
CKV

Problems
Suppose that we want to enhance the processor used for web serving. The new
processor is 10 times faster on computation in web serving application than the
original processor. Assuming that the original processor is busy with
computation 40 % of the time and is waiting for I/O 60% of the time. What is the
overall speedup gained by incorporating the enhancement?
CKV

Problems
your company's internal studies show that a single core system working at 2 GHz
is sufficient for the demand on your processing power of 10 W. However you are
exploring whether you could save power by using to cores.

Assume 80% of your application is parallelizable. To what value could you


decrease the frequency of a dual core system and get the same performance
as single core system ? 

Assume your application is 40% parallelizable and the voltage may be


decreased linearly with the frequency. How much dynamic power would be
required for a dual core system?
CKV

Thank You for Attending

You might also like