E2011: Theoretical fundamentals of computer science
Topic 4: Introduction to computer architectures
Vlad Popovici, Ph.D.
Fac. of Science - RECETOX
Outline
1 Introduction
2 A bit of computer architecture
Central processing unit
Memory
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a2 / 37
Motivation
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a3 / 37
An abstract view of a computer system
Physics: electrons
Transistors, diodes -
devices
Analog circuits:
amplifiers, filters
Logic: adders,
memory
Digital circuits: gates
Microarchitecture:
data paths,
controllers
Architecture:
instructions,
registers
Operating system:
device drivers,
kernels
Application
software: programs
Abstraction
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a4 / 37
Another view
Hardware
System software
Applications
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a5 / 37
Eight great ideas in computer design
(from Patterson and Hennessy’s “Computer Organization and Design”)
1 design for Moore’s Law
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a6 / 37
Eight great ideas in computer design
(from Patterson and Hennessy’s “Computer Organization and Design”)
1 design for Moore’s Law
2 use abstraction to simplify design
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a6 / 37
Eight great ideas in computer design
(from Patterson and Hennessy’s “Computer Organization and Design”)
1 design for Moore’s Law
2 use abstraction to simplify design
3 make the common case fast
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a6 / 37
Eight great ideas in computer design
(from Patterson and Hennessy’s “Computer Organization and Design”)
1 design for Moore’s Law
2 use abstraction to simplify design
3 make the common case fast
4 performance via parallelism
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a6 / 37
Eight great ideas in computer design
(from Patterson and Hennessy’s “Computer Organization and Design”)
1 design for Moore’s Law
2 use abstraction to simplify design
3 make the common case fast
4 performance via parallelism
5 performance via pipelining
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a6 / 37
Eight great ideas in computer design
(from Patterson and Hennessy’s “Computer Organization and Design”)
1 design for Moore’s Law
2 use abstraction to simplify design
3 make the common case fast
4 performance via parallelism
5 performance via pipelining
6 performance via prediction
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a6 / 37
Eight great ideas in computer design
(from Patterson and Hennessy’s “Computer Organization and Design”)
1 design for Moore’s Law
2 use abstraction to simplify design
3 make the common case fast
4 performance via parallelism
5 performance via pipelining
6 performance via prediction
7 hierarchy of memories
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a6 / 37
Eight great ideas in computer design
(from Patterson and Hennessy’s “Computer Organization and Design”)
1 design for Moore’s Law
2 use abstraction to simplify design
3 make the common case fast
4 performance via parallelism
5 performance via pipelining
6 performance via prediction
7 hierarchy of memories
8 dependability via redundancy
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a6 / 37
Moore’s law
The number of transistors in cost-effective integrated circuit double every
18-24 months.
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a7 / 37
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a8 / 37
Chip manufacturing process
(from Patterson and Hennessy’s “Computer
Organization and Design”)
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a9 / 37
Performance
what is the performance of a computer?
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a10 / 37
Performance
what is the performance of a computer?
response time vs throughput
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a10 / 37
Performance
what is the performance of a computer?
response time vs throughput
hardware vs software performance
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a10 / 37
Performance
what is the performance of a computer?
response time vs throughput
hardware vs software performance
energy per instruction
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a10 / 37
Performance
what is the performance of a computer?
response time vs throughput
hardware vs software performance
energy per instruction
measuring performance
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a10 / 37
(from Patterson and Hennessy’s “Computer Organization and Design”)
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a11 / 37
General architecture
In a very simplistic view,
Computer = Central Processing Unit + Memory
CPU
Data and
Instructions
Memory
CPU
Data
Memory
Instructions
Memory
von Neumann
architecture
Harvard
architecture
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a12 / 37
Central Processing Unit (CPU)
Control Unit
Arithmetic and Logic Unit (ALU)
C P U
Input
device
Output
device
Memory
unit
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a13 / 37
Central Processing Unit (CPU)
CPU executes instructions read from memory
instructions for loading and storing values
instructions that operate on values from registers, e.g. additions,
bitwise operations, math functions etc.
branching instructions
etc
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a14 / 37
CPU
Registers: internal (to CPU) memory cells used
MEMORY
INSTRUCTIONS
R2=LOAD 0x100
R1=100
0x100 | 10
0x090 | 0
0x120 | 0
CPU
0x110 | 110
R3=ADD R1,R2
STORE 0x110=R3
REGISTERS
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a15 / 37
Speed, clock, cycles
internal clock: used to maintain
synchronicity of th operations
the frequency of the clock (in
MHz, or GHz nowadays) gives
the speed of the CPU: one
operation may start on each tick
One cycle
Amplitude
One cycle
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a16 / 37
Instruction cycle
Main steps in executing an instruction
fetch: read instruction from memory
decode: figure out what to do
execute: take values from register and execute instruction
store: save the result in a register
Fetch Decode Execute Store
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a17 / 37
CPU: more details
Store
Load
FP
* /
+ -
FP
Decode Instruction
Floating Point Register File
AGU ALU
SSE/MMX (etc)
program code
Integer Register File
Cache
CPU
RAM
register: fast internal
storage; small - several
bytes per register
register file: the set of
similar registers within
CPU
register are specialized:
storing integer, floating
point, instructions,
addresses etc
AGU: address generation
unit - handles data
access
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a18 / 37
CPU: pipelines
Fetch Decode Execute Store
Fetch Decode Execute Store
Fetch Decode Execute Store
Fetch Decode Execute Store
Fetch Decode Execute Store
t1 t2 t3 t4 t5 t6 t7
Instruction 1
Instruction 2
Instruction 3
Instruction 4
Instruction 5
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a19 / 37
CPU: CISC vs RISC
CISC: Complex Instruction Set
Computer
the original ISA
one instruction may take several
cycles
emphasizes hardware over
software
complex instructions (e.g
memory-to-memory
LOAD/STORE)
shorter programs
high cycles per second
RISC: Reduced Instruction Set
Computer
improvement on CISC
one clock-cycle per instruction
emphasis on software
register-to-register
LOAD/STORE
uses many internal registers
low cycles per second
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a20 / 37
CPU: CISC vs RISC
Example: compute A × B. Assume A is stored at memory location 1200,
and B at 1201, respectively.
The following instruction(s) performs the multiplication and stores the
result at the first memory location.
CISC
MUL 1200,1201
RISC
Load A, 1200
Load B, 1201
Mul A, B
Store 1200, A
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a21 / 37
CPU: multilevel cache
cache: fast memory closer to CPU
improves data access speed by reducting emphmiss penalty
CPU
L1cache
L2cache
L3cache
Main
memory
(RAM)
Price
Storage space
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a22 / 37
Moving bits and bytes - data buses
a (computer) bus refers to
hardware and protocols for
transferring data
internal buses: data (memory)
bus, system bus, control bus, etc
external (expansion) buses:
connects devices to computer
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a23 / 37
Parallelism
SMP: symmetric multiprocessor systems
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a24 / 37
Parallelism
SMP: symmetric multiprocessor systems
Advantages:
increased throughput
redundancy, hency reliability
easy configuration.
more processes executing at
same time: MultiProcessing.
Drawbacks:
increased traffic over bus, longer
distances between two CPUs
risk of bottlenecks on shared
resources
coordination becomes much
more complex
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a25 / 37
Parallelism
Multicore
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a26 / 37
Parallelism
Multicore - example OpenSPARC (Sun Microsystems)
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a27 / 37
Parallelism
Multicore
Advantages:
run instructions in parallel on
different cores
usually use a single die, or onto
multiple dies but in single chip
package
more energy efficient: higher
performance at lower energy
less traffic, shorter distances
than SMP
Drawbacks:
overhead in writing specific code
dual-core processor does not
work at 2× speed of single
processor, but 60% − 80% more
speed
some operating systems still not
exploit the multicore
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a28 / 37
Memory organization
Computer = Central Processing Unit + MemoryMain Memory in the System
CORE 1
L2CACHE0
SHAREDL3CACHE
DRAMINTERFACE
CORE 0
CORE 2 CORE 3
L2CACHE1
L2CACHE2
L2CACHE3
DRAMBANKS
DRAM MEMORY
CONTROLLER
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a29 / 37
Wishes:
instantaneous access to any bit (0-latency)
infinite capacity
cheap (i.e. 0$)
infinite bandwidth
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a30 / 37
Wishes:
instantaneous access to any bit (0-latency)
infinite capacity
cheap (i.e. 0$)
infinite bandwidth
Reality:
larger memory is slower: more time to locate the desired position
faster memory is more expensive (SRAM vs DRAM)
larger bandwidth is more expensive
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a30 / 37
Memory technology
SRAM
Static Random Access Memory
per bit: 2 transistors for access,
4 transistors for storage
it keeps state as long as the
power is on
DRAM
Dynamic Random Access
Memory
per bit: 1 capacitor, 1 access
transistor
loses charge over time → needs
refresh cycles
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a31 / 37
level 0 (volatile): CPU registers:
data for instructions, etc
level 1 (volatile): L1 cache:
SRAM, separate data and
instruction space, KBs/core
level 2 (volatile): L2/3 cache:
SRAM, normally within the
same chip as CPU, MBs/core
level 3 (volatile): main memory:
usually DRAM; tens GBs (less
often hundreds GBs or 1TB); in
embedded devices could be
SRAM (KBs-MBs in size)
level 4 (permanent): disks, SSD
- TBs in size
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a32 / 37
Memory - other storage media
Floppy disks - now mostly extinct
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a33 / 37
Magnetic tapes - still relevant since 50s...
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a34 / 37
Magnetic tapes - still relevant since 50s...
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a35 / 37
Flash memory
non-volatile electronic memory
that can be electrically
reprogrammed
based on NAND or NOR gates
limited number of write/erase
cycles
data degradation over time
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a36 / 37
Questions?
Vlad Popovici, Ph.D. (Fac. of Science - RECETOX)E2011: Theoretical fundamentals of computer science Topic 4: Introduction to computer a37 / 37