01 Intro

Computer Architecture, Lecture 1:
What is Computer Architecture?
Hossam A. H. Fahmy
Cairo University
Electronics and Communications Engineering
1 / 17
Building at the nanoscale
According to the 1913 Webster, architecture is:

the art or science of building;. . . or
construction, in a more general sense.
Now, we get:
$ dict architecture
From wn [wn]:
architecture
..
.
4: (computer science) the structure and organization of a

computer’s hardware or system software; "the architecture
of a computer’s system software" [syn: {computer
architecture}]
2 / 17
The two architectures

Goals function, cost, safety, ease of 



construction, energy efficiency, fast




build time, aesthetics, . . .


⇒ Plans
Materials steel, concrete, brick, wood, glass, . . . 


Buildings houses, mosques, offices, museums,




schools, . . .




Goals function, performance, reliability, 

cost, ease of manufacture, energy 



effieciency, time to market, . . .





Technology logic gates, SRAM, DRAM, circuit

⇒ Plans
techniques, packaging, magnetic 

storage, . . .





Computers PCs, servers, PDAs, supercomputers, 




embedded, . . .
3 / 17
What are the differences?
There are at least three important differences:

1 the age (thousands of years versus about 70),
2 the rate of change, and
3 the automated mass production.
4 / 17
Where does computer architecture fit?
Application Software
Operating systems, Compilers, Networking software
Computer Architecture
Digital design
Circuits, Wires, Devices, Network hardware
The computer architecture interacts with many fields. It cannot

be studied in vaccum.
5 / 17
Design goals
Domain Is it a general purpose processor or domain specific

(Graphics, Neural network, Encryption, . . . )?
Functional Should be correct! What functions should it support?
Reliable A spacecraft is different from a PC. Is it really?
Performance It is not just the frequency but the speed of real tasks.
You cannot please everyone all the time.
Low cost design cost (how big are the teams? How long do they
take?), manufacturing cost, testing cost, . . .
Energy efficiency this is the “running cost”. Energy is drawn from
various sources. The cooling is a big issue.
Your job as an architect is to balance all of that given the
technologies available and the specific target computer.
The goals, the technologies, and the targets are all changing!
6 / 17
Computer architecture: Structure
Within the processor: Registers, Operational Units (Integer,

Floating Point, special purpose, . . . )
Outside the processor: Memory, I/O, . . .
Instruction Set Architecture (ISA): What is the best for the
target application? Why?
Examples: Sun SPARC, MIPS, Intel x86 (IA32), IBM S/390.
Defines: data (types, storage, and addressing modes),
instruction (operation code) set, and instruction formats.
7 / 17
Computer architecture: Organization
Within the processor: Pipeline(s), Control Unit, Instruction

Cache, Data Cache, Branch Prediction, . . .
Outside the processor: Secondary Caches, Memory Interleaving,
Redundant Disk Arrays, Multi-Processors, . . .
Which implementation is better? How do you define better?
We always optimize according to some purpose (application)
that sets the conditions of the problem.
8 / 17
Same architecture and multiple circuits
Realization of the computer: This is the physical fabrication

and assembly.
It is possible to implement an architecture
(structure and organization) using
the 130nm CMOS technology or the 5nm
technology,
4 metalization layers or 11 metalization layers,
Aluminum or Copper for the wires, . . .
9 / 17
ISA types
The addressing modes define some types:

Load/Store Only the LOAD and STORE instructions refer to data
in the memory. The ALU instructions operate on
data in registers. Used by many processors:
PowerPC, MIPS, HPPA, SPARC.
R/M The ALU instructions have one source or the
destination in memory. Used in mainframes and
older microprocessors: IBM S/390, Intel x86
(IA32).
R+M The ALU instructions may have any (or all) the
arguments in memory. Not commonly used now
but was in the Dec VAX machines.
Which type has a variable instruction size? Why? Is this good?
10 / 17
Data types
1 Integers. How do we represent negative numbers?

2 Floating point. Why not fixed?
3 Decimal digits. Who needs those? Why not floating
decimal?
4 Characters. In what encoding? Is it portable?
5 Bit strings. Why use individual bits?
What is the size of each?
11 / 17
Operation types
In a program, we operate on the data and depending on the

results we choose among the alternative paths in the algorithm.
The program interacts with the world to get the initial data and
produce the final result.
1 Arithmetic or Logical operations handle integer, floating point,
decimal, or binary data. Can we operate on characters?
2 Comparison instructions check a condition and produce a TRUE
or FALSE result.
3 Control transfer instructions branch (or jump) to another part of
the program.
4 Load and Store instructions deal with the memory while I/O
instructions interact with other devices. Must they be separate?
12 / 17
Beyond the basics
Computers serve multiple programs and users simultaneously:

There are instructions to ask for services from another
program module or from the operating system and then to
return back. How do we pass the arguments and the
results?
To provide some security measures, the instructions that
handle critical tasks are used only by the OS. The
processor may be in different modes (for example a user
mode or a system mode).
Usually a Program Status Word (PSW) holds many pieces of
information regarding the state of the processor including:
condition results, user id, current instruction address,
user/system mode, enable/disable traps, . . .
13 / 17
The journey from the processor to the memory
There are three distinct levels:

1 The program sees an address space provided at the user level.
This is usually a flat linear space going from zero to the
maximum memory location allowed for a process.
2 The system sees a full system address space. To accomodate
multiple users and programs, the system relocates the programs
and provides protection against unauthorized memory access.
3 The hardware deals with the reality. With the use of caches and
virtual memory, the hardware resolves the virtual address to get
the required information from its actual location. The primary
concern is to shorten the average access time.
As a computer architect, you design whatever is required at
each level. What do you think is needed?
14 / 17
Moore’s law
“Cramming More Components onto Integrated Circuits” by

Gordon E. Moore, Electronics, 1965.
Observation: transistor density doubles annually. He was slightly
off. Since then the density doubled almost every 18 months (he
had only four data points).
Corollaries:
the cost per transistor halves at the same rate,
the speed increases with scaling, and
the performance doubles almost every two years.
Do the periods (creation–2022) and (2022–2024) have the same
absolute increase in computing power per processor core?
Ans: This was true in previous periods. Now, parallel cores are
the trend.
15 / 17
A look at the trends
Year Model CPU cores GPU Neural Memory Transistors

(Perf+Eff) cores cores count
2020 M1 8 8 16 8-16 GB 16 G
(4+4)
2021 M1 Pro 10 16 16 16-32 GB 34 G

(8+2)
2021 M1 Max 10 32 16 32-64 GB 57 G

(8+2)
2022 M1 Ultra 20 64 32 64-128 GB 114 G

(16+4)
16 / 17
Key issues
Current architectures have multiple pipelines and caches.

They include several cores per chip.
It is important to evaluate the various options that become
available with improvements in technology.
A good (cost-effective) allocation of resources is necessary
for the design’s success in the market.
A high production volume is critical to recover costs.
In this class, you will learn to analyze, model, and look at
design targets. Are you up to the challenge of keeping (or
improving!) the historical trends?
17 / 17

01 Intro

Uploaded by

Copyright:

Available Formats

01 Intro

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

01 Intro

Uploaded by

Copyright:

Available Formats

Computer Architecture, Lecture 1:

What is Computer Architecture?

According to the 1913 Webster, architecture is:

4: (computer science) the structure and organization of a

There are at least three important differences:

Operating systems, Compilers, Networking software

Circuits, Wires, Devices, Network hardware

The computer architecture interacts with many fields. It cannot

Domain Is it a general purpose processor or domain specific

Within the processor: Registers, Operational Units (Integer,

Within the processor: Pipeline(s), Control Unit, Instruction

Realization of the computer: This is the physical fabrication

The addressing modes define some types:

1 Integers. How do we represent negative numbers?

In a program, we operate on the data and depending on the

Computers serve multiple programs and users simultaneously:

There are three distinct levels:

“Cramming More Components onto Integrated Circuits” by

Year Model CPU cores GPU Neural Memory Transistors

2021 M1 Pro 10 16 16 16-32 GB 34 G

2021 M1 Max 10 32 16 32-64 GB 57 G

2022 M1 Ultra 20 64 32 64-128 GB 114 G

Current architectures have multiple pipelines and caches.

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.