Learning Outcomes
SLO | After this lecture you can… | CLO | Bloom |
|---|---|---|---|
SLO 1.1 | Differentiate between computer architecture, organization and hardware using real examples. | CLO1 | C2 Understand |
SLO 1.2 | Classify computers into five classes and state the main design concern of each class. | CLO1 | C2 Understand |
SLO 1.3 | Identify the four ways hardware exploits parallelism and where each appears in this course. | CLO1 | C1 Remember |
SLO 1.4 | Explain the task of a computer designer: functional requirements, design goals and trade-offs. | CLO1 | C2 Understand |
CLO1: Describe the principles of computer design, including technology, cost and dependability trends.
§1 Starting Point and Course Map
§1.1 What you already know
Slide L01-S02
In Computer Organization and Assembly Language you learned what a computer is made of: registers (very small, very fast storage inside the CPU), the clock (a regular electric pulse that sets the speed of each step), cache (small fast memory near the CPU) and instructions such as ADD, LOAD, STORE and JUMP. That course answered what. This course answers why computers are built this way, and how designers make them faster, cheaper and more energy-efficient (using less electrical energy for the same work).
Quick check: A 2 GHz clock ticks 2 × 10⁹ times per second, so one clock cycle = 1 / (2 × 10⁹) s = 0.5 ns (nanoseconds). |
§1.2 Course map and assessment
Slides L01-S05, L01-S06
Unit | Lectures | Main question |
|---|---|---|
1 Quantitative principles | L1-L6 | How do we measure and compare computers? |
2 Instruction sets | L7-L12 | What is the 'language' of the CPU and how is it encoded? |
3 Pipelining and ILP | L13-L22 | How does one core run many instructions at the same time? |
4 Thread-level parallelism | L23-L29 | How do many cores share memory correctly? |
5 Data-level parallelism | L30-L32 | How do vectors, SIMD and GPUs process many data items at once? |
Labs | P1-P32 | Two 50-minute MIPS assembly sessions per week on the MARS simulator |
Marks: Mid-term 18 (30%, week 9) + Sessional 6 (10%) + Final 36 (60%) = 60. Main book: Hennessy and Patterson, Computer Architecture: A Quantitative Approach (2012).
§2 What Is Computer Architecture?
§2.1 Three levels of a computer design
Slide L01-S08
Level | Meaning | Examples |
|---|---|---|
Instruction Set Architecture (ISA) | The part of the computer visible to the programmer and compiler: instructions, registers, data types, addressing modes and memory model. It is the contract between software and hardware. | x86-64, ARMv8, MIPS32, RISC-V |
Organization (microarchitecture) | How the ISA is built: pipeline depth, cache sizes, number of cores, memory system, buses. | Intel Core i7 and AMD Ryzen 7: same ISA, different pipelines and caches |
Hardware | Detailed logic design and chip technology: transistor size, clock rate, packaging (the chip's outer case). | The same design made at 7 nm and at 5 nm |
Older books used 'architecture' to mean only the ISA. Today the word covers all three levels, because a designer must choose all three together to meet goals of performance, cost, energy and reliability (Hennessy and Patterson, §1.3).
§2.2 Real-world example: driving different cars
Slide L01-S09
A person with a driving licence can drive a Suzuki Alto and a Toyota Corolla because the steering, pedals and gear pattern are the same. These controls are like the ISA. The engines under the bonnet are very different (organization), and the exact parts and factory are different again (hardware). In the same way, a Windows program (.exe file) runs on both an Intel Core i7 and an AMD Ryzen 7 because both implement x86-64. This is called binary compatibility (old machine-code files still run on new hardware). It is the reason an ISA can live for decades: x86 has been extended since 1978 but old programs still run.
§2.3 Layers of abstraction
Slide L01-S10
Abstraction means hiding the details of a lower layer so that the upper layer can work simply. A computer system is a stack of layers:
Layer | Example |
|---|---|
Application | WhatsApp, MS Word, a Python script |
Operating system and compiler | Windows, Android, gcc |
Instruction set architecture | the software/hardware contract |
Organization (microarchitecture) | pipeline, caches, cores |
Logic and circuits | gates, adders, flip-flops |
Devices | transistors (billions of tiny switches) |
This course works mainly at the two layers next to the ISA: how compilers use it (Unit 2) and how the organization runs it quickly (Units 3-5).
§3 Classes of Computers
§3.1 The five classes
Slide L01-S12
Class | System price | Processor price | Critical design concerns |
|---|---|---|---|
Personal mobile device (PMD) | $100-$1,000 | $10-$100 | Cost, energy, media performance, responsiveness |
Desktop / laptop | $300-$2,500 | $50-$500 | Price-performance, energy, graphics performance |
Server | $5,000-$10,000,000 | $200-$2,000 | Throughput, availability, scalability, energy |
Cluster / warehouse-scale computer (WSC) | $100,000-$200,000,000 | $50-$250 | Price-performance, throughput, energy proportionality |
Embedded | $10-$100,000 | $0.01-$100 | Price, energy, application-specific performance |
Source: Hennessy and Patterson (2012), Figure 1.2.
Throughput = work done per unit time (e.g., transactions per second). Scalability = the ability to grow by adding more processors, memory or disks. Energy proportionality = the power used grows in proportion to the work done, so an idle server uses little power.
§3.2 Examples around us
Slide L01-S13
Class | Local example | Why its concern matters |
|---|---|---|
PMD | Infinix or Samsung phone | Battery must last a full day; apps must respond quickly |
Desktop | UAF lab PC, gaming laptop | Best speed for the money |
Server | NADRA citizen database, bank core-banking servers | If a bank app is down on salary day, millions are affected |
WSC | Google or Microsoft Azure data centers | Electricity and cooling are a major running cost |
Embedded | Safe City camera, car engine control unit (ECU), microwave oven | Sold in millions, so every rupee and every milliwatt counts |
§3.3 Worked example: availability and downtime
Slide L01-S14
Availability is the fraction of time a system works correctly. One year has 365 × 24 = 8,760 hours.
Availability | Calculation | Downtime per year |
|---|---|---|
99.9% | 0.001 × 8,760 h | 8.76 hours |
99.99% | 0.0001 × 8,760 h = 0.876 h | 52.56 minutes |
99.999% ('five nines') | 0.00001 × 8,760 h = 0.0876 h | 5.26 minutes |
Each extra 'nine' needs more spare hardware (backup servers, power supplies, a second data center), so it costs more money. A bank or NADRA chooses more nines; a student's personal website does not need them. Dependability (MTTF, MTTR) is covered in Lecture 3.
§4 Kinds of Parallelism
Slide L01-S16
Parallelism means doing several things at the same time. Applications contain two basic kinds:
- Data-level parallelism (DLP): many data items can be processed together (e.g., brighten every pixel of a photo).
- Task-level parallelism: many independent tasks can run together (e.g., many students using a website).
Hardware exploits these in four ways, and each is a unit of this course:
Hardware technique | What it does | Where in CS-704 |
|---|---|---|
Instruction-level parallelism (ILP) | Overlaps instructions of one program: pipelining, speculative execution | Unit 3 (L13-L22) |
Vector architectures, SIMD, GPUs | One instruction works on many data items (DLP) | Unit 5 (L30-L32) |
Thread-level parallelism (TLP) | Many threads run on many cores that share memory | Unit 4 (L23-L29) |
Request-level parallelism (RLP) | Many independent user requests served by many servers | Data centers (overview) |
Example: When 5,000 students check results on the UAF website at the same time, each request is independent, so the servers use request-level parallelism. |
§5 The Task of a Computer Designer
§5.1 What a designer does
Slide L01-S18
The designer must decide which attributes are important for a new computer, then design it to maximize performance and energy efficiency while staying within limits on cost, power and availability (Hennessy and Patterson, §1.3). The work is a loop:
- 1. Requirements: Which market? Which software must run?
- 2. Goals: Targets for performance, energy, cost, availability.
- 3. Design: Choose the ISA, organization and hardware.
- 4. Measure: Benchmarks, simulation, cost models (Lectures 4-6).
- 5. Trade-off: Improve the common case, then repeat.
The design must also last: a chip takes years to design and sells for years, so the designer must plan for technology trends (Lecture 2).
§5.2 Functional requirements
Slide L01-S19
Requirement | Choices | What it means for the design |
|---|---|---|
Application area | General-purpose, scientific, commercial, embedded | Balanced speed vs high floating point vs high availability vs low cost |
Software compatibility | Language level vs object-code (binary) level | Binary level fixes the ISA but protects existing software |
Operating system needs | Address space size, memory management, protection | e.g., 64-bit addresses, page tables, user/kernel modes |
Standards | Floating point, I/O, networks, languages | IEEE 754, USB, PCIe, Ethernet, Wi-Fi, C/C++ |
Source: adapted from Hennessy and Patterson (2012), Figure 1.7. Object code is the machine-language file that the CPU runs directly. Language-level compatibility means programs must be recompiled for the new machine, which gives the designer more freedom.
§5.3 Trade-offs: one problem, two designs
Slide L01-S20
A trade-off means giving up some of one thing to get more of another.
Decision | Chip for a Rs 30,000 phone | Chip for a bank server |
|---|---|---|
Most important goal | Battery life (energy) | Availability and throughput |
Power budget | About 2-5 watts | 100-300 watts |
Cores | Few big + few small cores | Many identical large cores |
Memory | Small, shared with graphics | Very large, error-correcting (ECC) |
Price of processor | Low | High, but a small part of the system cost |
Lesson: There is no single best computer, only the best computer for a given set of requirements. |
§6 Common Mistakes
Slide L01-S22
Wrong idea | Correct idea |
|---|---|
Architecture means only the instruction set. | Architecture today covers the ISA, organization and hardware, designed together. |
A faster clock always means a faster computer. | Speed also depends on instructions per program and cycles per instruction (Lecture 4). |
Servers and phones need the same design. | Each class has different critical concerns: energy for phones, availability for servers. |
Same ISA means same performance. | Core i7 and Ryzen share x86-64 but have different organizations, so speed differs. |
Word Bank
Word | Simple meaning | Real-world example |
|---|---|---|
Abstraction | Hiding lower-level details | You drive a car without knowing how the engine burns fuel |
Availability | Fraction of time a system works | A bank app that is up 99.99% of the year |
Binary compatibility | Old machine-code programs run on new hardware | A 2010 Windows game still runs on a 2026 laptop |
Embedded computer | A computer hidden inside another product | The chip in a microwave or a car |
Energy efficiency | Doing the same work with less energy | A phone that lasts all day on one charge |
ISA | The instructions and registers the software can see | x86-64, ARMv8, MIPS32 |
Microarchitecture | How an ISA is built inside the chip | Pipeline and cache design of a Ryzen chip |
Parallelism | Doing many things at the same time | Many cashiers serving customers at Imtiaz store |
Scalability | Ability to grow by adding more parts | A bank adds servers as customers grow |
Throughput | Work done per unit time | Number of transactions a server handles per second |
Trade-off | Giving up one thing to gain another | A bigger battery makes the phone heavier |
Self-Check Questions
- Define ISA, organization and hardware. Give one example of each. (§2.1)
- Why can a Windows .exe file run on both Intel and AMD processors? (§2.2)
- Explain abstraction with a non-computer example. (§2.3)
- Name the five classes of computers and one critical design concern of each. (§3.1)
- Why does a WSC care about energy proportionality? (§3.1)
- Calculate the downtime per year for 99.95% availability. (§3.3)
- Name the four ways hardware exploits parallelism and the unit of this course that covers each. (§4)
- State the task of a computer designer in one sentence. (§5.1)
- What is the difference between language-level and object-code compatibility? (§5.2)
- Compare the design goals of a phone chip and a server chip. (§5.3)
Answer to the downtime question: (1 - 0.9995) × 8,760 = 0.0005 × 8,760 = 4.38 hours per year.
Further Reading
- Hennessy and Patterson (2012): Chapter 1, sections 1.1-1.3 (pages 2-17).
- Ledin (2020): Chapter 1, for the instruction set of x86, ARM and RISC-V in today's phones, PCs and cloud servers.