Upload
michael-richards
View
220
Download
0
Tags:
Embed Size (px)
Citation preview
Outline
Power & Cost Performance Performance measurement Amdahl's Law Benchmarks
History
1. “Big Iron” Computers: Used vacuum tubes, electric relays and bulk magnetic
storage devices. No microprocessors. No memory.Example: ENIAC (1945), IBM Mark 1 (1944)
History
Von Newmann:Invented EDSAC (1949).First Stored Program Computer. Uses Memory.
Importance: We are still using the same basic design.
Computer Components
Processor (CPU)
MemoryControl
Output
Input
keyboardMouseDisk
. . .
PrinterScreen
Disk. . .
Computer Components
Datapath of a von Newman machine
ALU
OP1 + OP2
Op1 Op2
OP1 + OP2...
Op1Op2
General-purpose Registers
ALU i/p registers
ALU o/p register
Bus
Computer Components
Processor(CPU): Active part of the motherboard Performs calculations & activates devices Gets instruction & data from memory Components are connected via Buses
Bus: Collection of parallel wires Transmits data, instructions, or control signals
Motherboard Physical chips for I/O connections, memory, & CPU
Computer Components
CPU consists of Datapath (ALU+ Registers):
Performs arithmetic & logical operations Control (CU):
Controls the data path, memory, & I/O devices
Sends signals that determine operations of datapath, memory, input & output
Technology Change
Technology changes rapidly HW
Vacuum tubes: Electron emitting devices Transistors: On-off switches controlled by electricity Integrated Circuits( IC/ Chips): Combines thousands of
transistors Very Large-Scale Integration( VLSI): Combines millions of
transistors What next?
SW Machine language: Zeros and ones Assembly language: Mnemonics High-Level Languages: English-like Artificial Intelligence languages: Functions & logic
predicates Object-Oriented Programming: Objects & operations on
objects
Moore’s Prediction
Moore’s Law:
A new generation of memory chips is introduced every 3 years Each new generation has 4 times as much memory as its predecessor
Computer technology doubles every 1.5 years:
Example: DRAM capacity
Kb
it c a
pa
city
Year of introduction1992
100,000
10,000
1000
100
1019901988198619841982198019781976
16M
4M
1M
256K
16K
64K
1994 1996
64M
Technology => dramatic change
Processor logic capacity: about 30% per year clock rate: about 20% per year
Memory DRAM capacity: about 60% per year (4x every 3 years) Memory speed: about 10% per year Cost per bit: improves about 25% per year
Disk capacity: about 60% per year
Question: Does every thing look OK?
Software Evolution.
Machine language Assembly language High-level languages Subroutine libraries
There is a large gap between what is convenient for computers & what is convenient for humans
Translation/Interpretation is needed between both
Language Evolution
swap (int v[], int k){ int temp temp = v[k]; v[k] = v[k+1]; v[k+1] = temp;}
00000000101000010000000000011000000000001000111000011000001000011000110001100010000000000000000010001100111100100000000000000100101011001111001000000000000000001010110001100010000000000000010000000011111000000000000000001000
swap: muli $2, $5, 4 add $2, $4, $2 lw $15, 0($2) lw $18, 4($2) sw $18, 0($2) sw $15, 4($2) jr $31
High-level language program (in C)
Assembly language program (for MIPS)
Binary machine language program (for MIPS)
HW - SW Components
Hardware Memory components Registers Register file memory Disks Functional components Adder, multiplier, dividers, . . . Comparators Control signals
Software Data Simple
• Characters• Integers• Floating-point• Pointers
Structured• Arrays• Structures ( records) Instructions
Data transfer Arithmetic Shift Control flow Comparison . . .
Things You Will Learn
Assembly language introduction/Review
How to analyze program performance
How to design processor components
How to enhance processors performance (caches, pipelines, parallel processors, multiprocessors)
The Processor Chip
Processor Chip Major Blocks
• Example: Intel Pentium • Area: 91 mm2
• ~ 3.3 million transistors ( 1 million for cache memory)
Data cache
ControlBranch
Instruction cache
Bus Integer data-path
Floating-point datapa
th
Memory
Categories Volatile memory
Loses information when power is switched-off Non-volatile memory
Keeps information when power is switched-off Types
Cache: Volatile Fast but expensive Smaller capacity Placed closer to the processor
Main memory Volatile Less expensive More capacity
Secondary memory Nonvolatile Low cost Very slow Unlimited capacity
Input-Output (I/O)
I/O devices have the hardest organization Wide range of speeds
Graphics vs. keyboard Wide range of requirements
SpeedStandardCost . . .
Least amount of research done in this area
Our Primary Focus
The processor (datapath and control) Implemented using millions of transistors Impossible to understand by looking at each transistor We need abstraction
Hides lower-level details to offer simple model at higher level
Advantages• Intensive & thorough research into the depths • Reveals more information• Omits unneeded details• Helps us cope with complexity
Examples of abstraction:
• Language hierarchy• Instruction set architecture (ISA)
Instruction Set Architecture (ISA)
Instruction set:
Complete set of instructions used by a machine
ISA:
Abstract interface between the HW and lowest-level SW. It encompasses information needed to write machine-language programs including
Instructions
Memory size
Registers used
. . .
Instruction Set Architecture (ISA)
ISA is considered part of the SW Several implementations for the same ISA can exist Modern ISA’s:
80x86/Pentium/K6, PowerPC, DEC Alpha, MIPS, SPARC, HP We are going to study MIPS
Advantages: Different implementations of the same architecture Easier to change than HW Standardizes instructions, machine language bit patterns, etc.
Disadvantage: Sometimes prevents using new innovations
Instruction Set Architecture (ISA)
Fetch Instruction From Memory
Decode Instruction determine its size & action
Fetch Operand data
Execute instruction & compute results or status
Store Result in memory
Determine Next Instruction’s address
•Instruction Execution Cycle
What Should we Learn?
A specific ISA (MIPS)
Performance issues - vocabulary and motivation
Instruction-Level Parallelism
How to Use Pipelining to improve performance
Exploiting Instruction-Level Parallelism w/ Software Approach
Memory: caches and virtual memory
I/O
What is Expected From You?
• Read textbook & readings!• Be up-to-date! • Come back with your input & questions for discussion!• Appreciate and participate in teamwork!
Power?
Everything is done by tiny switches
Their charge represents logic values Changing charge energy Power energy over time Devices are non-ideal power heat Excess heat Circuits breakdown
Need to keep power within acceptable limitsNeed to keep power within acceptable limits
POWER in the real world
1
10
100
1000
W/c
m2
Integrated Circuits Costs
Die cost = Wafer cost
Dies per Wafer * Die yield
Dies per wafer = Wafer Area * Wafer diameter
Die Area (2 * die area)1/2
Die yield = Wafer yield * ( 1+ (Defects per unit area * Die area )/ ) - = 4.0
Percentage of good dies on a wafer
Integrated Circuits Costs-example
Find the die yield for a die with a defect density of 0.6 for dies with areas 1.0 and .49.
For the larger die: Die yield = (1+ (0.6*1)/4) -4 = 0.57
for the smaller die: Die yield = (1+ (0.6*0.49)/4) -4 = 0.75
Why?
IC cost = Die cost + Testing cost + Packaging cost
Final test yield
Packaging Cost: depends on pins, heat dissipation, ...
Other Costs
Chip Die Packaging Testing Total
386DX $4 $1 $4 $9
486DX2 $12 $11 $12 $35
PowerPC 601 $53 $3 $21 $77
HP PA 7100 $73 $35 $16 $124
DEC Alpha $149 $30 $23 $202
SuperSPARC $272 $20 $34 $326
Pentium $417 $19 $37 $473
System Cost: Workstation
System Subsystem% of total cost
Cabinet Sheet metal, plastic1%
Power supply, fans2%
Cables, nuts, bolts1%
(Subtotal)(4%)
Motherboard Processor6%
DRAM (64MB)36%
Video system14%
I/O system3%
Printed Circuit board1%
(Subtotal)(60%)
I/O Devices Keyboard, mouse1%
Monitor22%
Hard disk (1 GB)7%
Tape drive (DAT)6%
(Subtotal)(36%)
COST v. PRICE
ComponentCost
componentcost
Direct Costs
componentcost
direct costs
Gross Margin
componentcost
direct costs
gross margin
AverageDiscount
list price
avg. selling price
Input: chips, displays, ...
Making it: labor, scrap, returns, ...
Overhead: R&D, rent, marketing, profits, ...
Commission: channel profit, volume discounts,
+33%
+25–100%
+50–80%
(25–31%)
(33–45%)
(8–10%)
(33–14%)
(WS–PC)Q: What % of company incomeon Research and Development (R&D)?
Performance
Purchasing perspective given a collection of machines, which has the
best performance ? least cost ? best performance / cost ?
Design perspective faced with design options, which has the
best performance improvement ? least cost ? best performance / cost ?
Both require basis for comparison metric for evaluation
Our goal is to understand cost & performance implications of architectural choices
Two notions of “performance”
° Time to do the task (Execution Time)
– execution time, response time, latency
° Tasks per day, hour, week, sec, ns. ..
– throughput, bandwidth
Response time and throughput often are in opposition
DC to Paris
6.5 hours
3 hours
Plane
Boeing 747
BAD/Sud Concorde
Speed
610 mph
1350 mph
Passengers
470
132
Throughput
286,700
178,200
Which has higher performance?
Example
• Time of Concorde vs. Boeing 747?
• Concord is 1350 mph / 610 mph = 2.2 times faster
= 6.5 hours / 3 hours
• Throughput of Concorde vs. Boeing 747 ?
• Concord is 178,200 pmph / 286,700 pmph = 0.62 “times faster”
• Boeing is 286,700 pmph / 178,200 pmph = 1.6 “times faster”
• Boeing is 1.6 times (“60%”)faster in terms of throughput
• Concord is 2.2 times (“120%”) faster in terms of flying time
We will focus primarily on execution time for a single job
Definitions
Performance is in units of things-per-second bigger is better
If we are primarily concerned with response time performance(x) = 1
execution_time(x)
" X is n times faster than Y" means
Performance(X)
n = ----------------------
Performance(Y)
Performance measurement
How about collection of programs? Example:
Three machines: A, B and C. Two Programs: P1 and P2.
A B C W(1) W(2) W(3)
P1 1 10 20 .5 .9 .99
P2 1000 100 20 .5 .1 .01
W(1) 500.5 55 20
Arithmetic mean: Weight i * Time i
W(2) 91.9 18 20
W(3) 2 10 20
Performance measurement
Other option: Geometric Means (Self study pages 37-39 text book)
Metrics of performance
Compiler
Programming Language
Application
DatapathControl
Transistors Wires Pins
ISA
Function Units
(millions) of Instructions per second – MIPS(millions) of (F.P.) operations per second – MFLOP/s
Cycles per second (clock rate)
Megabytes per second
Answers per monthOperations per second
Relating Processor Metrics
CPU execution time = CPU clock cycles X clock cycle time
or CPU execution time = CPU clock cycles ÷ clock rate
CPU clock cycles= Instructions X avg. clock cycles per instr.
or CPI = CPU clock cycles÷ Instructions
CPI tells us something about the Instruction Set Architecture, the Implementation of that architecture, and the program measured
Aspects of CPU PerformanceCPU time = Seconds = Instructions x Cycles x
Seconds
Program Program Instruction Cycle
CPU time = Seconds = Instructions x Cycles x Seconds
Program Program Instruction Cycle
instr. count CPI clock rate
Program
Compiler
Instr. Set Arch.
Organization
Technology
Aspects of CPU PerformanceCPU time = Seconds = Instructions x Cycles x
Seconds
Program Program Instruction Cycle
CPU time = Seconds = Instructions x Cycles x Seconds
Program Program Instruction Cycle
instr count CPI clock rate
Program X
Compiler X (x)
Instr. Set. X X
Organization X X
Technology X
Organizational Trade-offs
Compiler
Programming Language
Application
DatapathControl
Transistors Wires Pins
ISA
Function Units
Instruction Mix
Cycle Time
CPI
CPI
CPU time = ClockCycleTime * SUM CPI * Ii = 1
n
ii
CPI = SUM CPI * F where F = I i = 1
n
i i i i
Instruction Count
"instruction frequency"
Invest Resources where time is Spent!
CPI = (CPU Time * Clock Rate) / Instruction Count = Clock Cycles / Instruction Count
“Average cycles per instruction”
Example (RISC processor)
Typical Mix
Base Machine (Reg / Reg)
Op Freq Cycles CPI(i) % Time
ALU 50% 1 .5 23%
Load 20% 5 1.0 45%
Store 10% 3 .3 14%
Branch 20% 2 .4 18%
2.2
How much faster would the machine be if a better data cachereduced the average load time to 2 cycles?
How does this compare with using branch prediction to shave a cycle off the branch time?
What if two ALU instructions could be executed at once?
Example (RISC processor)
Typical Mix
Base Machine (Reg / Reg)
Op Freq Cycles CPI(i) % Time
ALU 50% 1 .5 23%
Load 20% 5 1.0 45%
Store 10% 3 .3 14%
Branch 20% 2 .4 18%
2.2
How much faster would the machine be if:A) Loads took “0” cycles? B) Stores took “0” cycles?C) ALU ops took “0” cycles?D)Branches took “0” cycles?
MAKE THE COMMON CASE FAST
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 then,
ExTime(with E) = ((1-F) + F/S) X ExTime(without E)
Speedup(with E) = ExTime(without E) ÷ ((1-F) + F/S) X ExTime(without E)
Speedup(with E) =1/ ((1-F) + F/S)
Amdahl's Law-example
A new CPU makes Web serving 10 times faster. The old CPU spent 40% of the time on computation and 60% on waiting for I/O. What is the overall enhancement?
Fraction enhanced= 0.4
Speedup enhanced = 10
Speedup overall = 1 = 1.56
0.6 +0.4/10
Example from Quiz 1-2004
a)A program consists of 80% initialization code and of 20% code being the main iteration loop, which is run 1000 times. The total runtime of the program is 100 seconds. Calculate the fraction of the total run time needed for the initialization and the iteration. Which part would you optimize? B) The program should have a total run time of 60 seconds. How can this be achieved? (15 points)
Marketing Metrics
MIPS = Instruction Count / Time * 10^6
= Clock Rate / CPI * 10^6
•machines with different instruction sets ?
•programs with different instruction mixes ?
• dynamic frequency of instructions
• uncorrelated with performance
GFLOPS = FP Operations / Time * 10^9 playstation: 6.4 GFLOPS
•machine dependent
•often not where time is spent
Why Do Benchmarks? How we evaluate differences
Different systems Changes to a single system
Provide a target Benchmarks should represent large class of important
programs Improving benchmark performance should help many
programs For better or worse, benchmarks shape a field Good ones accelerate progress
good target for development Bad benchmarks hurt progress
help real programs v. sell machines/papers? Inventions that help real programs don’t help
benchmark
Basis of Evaluation
Actual Target Workload
Full Application Benchmarks
Small “Kernel” Benchmarks
Microbenchmarks
Cons
• representative• very specific• non-portable• difficult to run, or measure• hard to identify cause
• portable• widely used• improvements useful in reality
• easy to run, early in design cycle
• identify peak capability and potential bottlenecks
•less representative
• easy to “fool”
• “peak” may be a long way from application performance
Successful Benchmark: SPEC
1987 RISC industry mired in “bench marketing”:(“That is 8 MIPS machine, but they claim 10 MIPS!”)
EE Times + 5 companies band together to perform Systems Performance Evaluation Committee (SPEC) in 1988: Sun, MIPS, HP, Apollo, DEC
Create standard list of programs, inputs, reporting: some real programs, includes OS calls, some I/O
SPEC first round
First round 1989; 10 programs, single number to summarize performance
One program: 99% of time in single line of code New front-end compiler could improve dramatically
Benchmark
SPE
C P
erf
0
100
200
300
400
500
600
700
800
gcc
epre
sso
spic
e
doduc
nasa7
li
eqnto
tt
matr
ix300
fpppp
tom
catv
SPEC95
Eighteen application benchmarks (with inputs) reflecting a technical computing workload
Eight integer go, m88ksim, gcc, compress, li, ijpeg, perl, vortex
Ten floating-point intensive tomcatv, swim, su2cor, hydro2d, mgrid, applu,
turb3d, apsi, fppp, wave5 Must run with standard compiler flags
eliminate special undocumented incantations that may not even generate working code for real programs
Summary
Time is the measure of computer performance! Good products created when have:
Good benchmarks Good ways to summarize performance
If not good benchmarks and summary, then choice between improving product for real programs vs. improving product to get more sales=> sales almost always wins
Remember Amdahl’s Law: Speedup is limited by unimproved part of program
CPU time = Seconds = Instructions x Cycles x Seconds
Program Program Instruction Cycle
CPU time = Seconds = Instructions x Cycles x Seconds
Program Program Instruction Cycle
Readings & More…
Reminder:
READ:
TEXTBOOK: Chapter 1 pages 1 to 47 Moore paper (posted on course web site).