CENG 3420

Computer Organization & Design

Lecture 10: Performance
Bei Yu
CSE Department, CUHK
byu@cse.cuhk.edu.hk

(Textbook: Chapters 1.6 & 1.7)

Spring 2022
 Performance Metrics

Purchasing perspective
Given a collection of machines, which has the
• best performance ?
• least cost ?
• best cost/performance?

Design perspective
Faced with design options, which has the
• best performance improvement ?
• least cost ?
• best cost/performance?

2/21
 Performance Metrics

Both (purchasing & design) require

• basis for comparison
• metric for evaluation

Our Goal
Is to understand what factors in the architecture contribute to overall system performance
and the relative importance (and cost) of these factors

3/21
 Throughput v.s. Response Time

Response time (execution time)

• The time between the start and the completion of a task.
• Important to individual users

Throughput (bandwidth)
• The total amount of work done in a given time
• Important to data center managers

Will need different performance metrics as well as a different set of applications to

benchmark embedded and desktop computers, which are more focused on
response time, versus servers, which are more focused on throughput

4/21
Response Time Matters

Justin Rattner’s ISCA-08 Keynote (VP and CTO of Intel) 5/21

Defining (Speed) Performance

• To maximize performance, need to minimize execution time

1
performanceX =
execution_timeX

• If X is n times faster than Y, then

performanceX execution_timeY
= =n
performanceY execution_timeX

• Decreasing response time almost always improves throughput.

6/21
EX-1
If computer A runs a program in 10 seconds and computer B runs the same program in 15
seconds, how much faster is A than B?

7/21
Performance Factors

• CPU execution time (CPU time): time the CPU spends working on a task
• Does not include time waiting for I/O or running other programs

CPU execution time = # CPU clock cycles × clock cycle time

# CPU clock cycles
=
clock rate

Can improve performance by reducing

• Length of the clock cycle
• Number of clock cycles required for a program

8/21
 Review: Machine Clock Rate

Clock rate (clock cycles per second in MHz or GHz) is inverse of clock cycle time
(clock period)
1
CC =
CR

one  clock  p eriod

10  nsec clock  cycle    =>    100  MHz  clock  rate

5  nsec clock  cycle    =>    200  MHz  clock  rate
2  nsec clock  cycle    =>    500  MHz  clock  rate
1  nsec (10-­9)  clock  cycle      =>    1  GHz  (109)  clock  rate
500  psec clock  cycle    =>      2  GHz  clock  rate
250  psec clock  cycle    =>      4  GHz  clock  rate
200  psec clock  cycle    =>      5  GHz  clock  rate

9/21
EX-2: Improving Performance Example
A program runs on computer A with a 2 GHz clock in 10 seconds. What clock rate must a
computer B has to run this program in 6 seconds? Unfortunately, to accomplish this,
computer B will require 1.2 times as many clock cycles as computer A to run the program.

10/21
 Clock Cycles per Instruction

• Not all instructions take the same amount of time to execute

• One way to think about execution time is that it equals the number of instructions
executed multiplied by the average time per instruction
CPU clock cycles = # instruction × clock cycle per instruction

Clock cycles per instruction (CPI)

• The average number of clock cycles each instruction takes to execute
• A way to compare two different implementations of the same ISA

11/21
Effective (Average) CPI

n
X
CPIi × ICi
i=1

ICi : percentage of the number of instructions of class i executed

CPIi : (average) number of clock cycles per instruction for that instruction class
n: number of instruction classes

• Computing the overall effective CPI is done by looking at the different types of
instructions and their individual cycle counts and averaging
• The overall effective CPI varies by instruction mix
• A measure of the dynamic frequency of instructions across one or many programs

12/21
 Basic Performance Equation

CPU time = Instruction count × CPI × clock cycle

Instruction count × CPI
CPU time =
clock rate

Three key factors that affect performance

• Can measure the CPU execution time by running the program
• The clock rate is usually given
• Can measure overall instruction count by using profilers/ simulators without
knowing all of the implementation details

CPI varies by instruction type and ISA implementation for which we must know
the implementation details
13/21
EX-3: Using the Performance Equation
Computers A and B implement the same ISA. Computer A has a clock cycle time of 250 ps
and an effective CPI of 2.0 for some program and computer B has a clock cycle time of 500
ps and an effective CPI of 1.2 for the same program. Which computer is faster and by how
much?

14/21
Determinates of CPU Performance

CPU time = Instruction count × CPI × clock cycle

Instruction_ CPI clock_cycle

count
Algorithm

Programming  
language
Compiler

ISA

Core  
organization
Technology

15/21
Determinates of CPU Performance

CPU time = Instruction count × CPI × clock cycle

Instruction_ CPI clock_cycle

count
Algorithm
X X

Programming X X
language
Compiler X X

ISA
X X X

Core
X X
organization
Technology
X

15/21
EX-4
Op Freq CPIi Freq x CPIi
ALU 50% 1
Load 20% 5
Store 10% 3
Branch 20% 2

∑=

1 How much faster would the machine be if a better data cache reduced the average
load time to 2 cycles?
2 How does this compare with using branch prediction to shave a cycle off the branch
time?
3 What if two ALU instructions could be executed at once?

16/21
 Workloads and Benchmarks

Benchmarks
A set of programs that form a “workload” specifically chosen to measure performance

• SPEC (System Performance Evaluation Cooperative) creates standard sets of

benchmarks starting with SPEC89.
• The latest is SPEC CPU2006 which consists of 12 integer benchmarks (CINT2006) and
17 floating-point benchmarks (CFP2006).
• www.spec.org
• There are also benchmark collections for power workloads (SPECpower_ssj2008), for
mail workloads (SPECmail2008), for multimedia workloads (mediabench) ...

17/21
SPEC CINT2006 on Barcelona (CC = 0.4 × 109 )

Name ICx109 CPI ExTime RefTime SPEC

ratio
perl 2,1118 0.75 637 9,770 15.3
bzip2 2,389 0.85 817 9,650 11.8
gcc 1,050 1.72 724 8,050 11.1
mcf 336 10.00 1,345 9,120 6.8
go 1,658 1.09 721 10,490 14.6
hmmer 2,783 0.80 890 9,330 10.5
sjeng 2,176 0.96 837 12,100 14.5
libquantum 1,623 1.61 1,047 20,720 19.8
h264avc 3,102 0.80 993 22,130 22.3
omnetpp 587 2.94 690 6,250 9.1
astar 1,082 1.79 773 7,020 9.1
xalancbmk 1,058 2.70 1,143 6,900 6.0
Geometric Mean 11.7
18/21
 Comparing and Summarizing Performance

How to summarize performance with a single number?

• First the execution times are normalized given the “SPEC ratio” (bigger is faster, i.e.,
SPEC ratio is the inverse of execution time)
• SPEC ratios are “averaged” using the geometric mean (GM)

v
u n
uX
GM = n · t SPEC ratioi
i=1

Guiding principle – reproducibility

List everything another experimenter would need to duplicate the experiment: version of
the operating system, compiler settings, input set used, specific computer configuration
(clock rate, cache sizes and speed, memory size and speed, etc.)

19/21
Other Performance Metrics

Power Consumption
• Especially in the embedded market where battery life is important
• For power-limited applications, the most important metric is energy efficiency

20/21
 Highest Clock Rate of Intel Processors

What if the exponential increase had kept up? Why not?

• Due to process improvements
• Deeper pipeline
• Circuit design techniques
21/21