Computer Architecture

What is computer architecture?

• A set of rules and methods that describe functionality, organisation and implementation of computer systems

• Can be applied to many layers

  1. Processor architecture

  2. Memory architecture

  3. Instruction set architecture

  4. System level: how we link processors to I/O

What is the Von Neumann architecture?

Proposed by John Von Neumann in 1945

1. Memory: stores data and instructions

2. Control unit: contains instruction register and program counter

3. Processing unit: contains ALU and processor registers

4. I/O mechanisms

What is the CPU composed of?

Consists of:

1. ALU: arithmetic logic unit

2. CU: also contains high-speed cache memory

What are advantages and disadvantages of the Harvard architecture?

• Advantages:

  • Overcomes bottleneck issues with Von Neumann

  • Parallel access to instruction and data = faster

  • Better against cyber attacks

• Disadvantages:

  • Higher costs as it needs more memory since data and instruction data storage is seperate

• IMPORTANT

  1. Instructions and data memory is seperate

  2. Cannot change instruction memory at run-time

What is the modified Harvard architecture?

Improvement to the original Harvard version

• Still seperates instruction and data caches INTERNALLY, BUT a single unified main memory is still visible to users/programs

• Used in chips e.g. ARM9, MIPS, x86

What is the modern PC architecture?

• Divided into regions according to speed of operation

• Memory controller: northbridge

  • Manages communication between CPU and RAM

  • Is now inside the CPU, allows faster access and lower latency

• Input / Output controller: southbridge

  • Manages I/O devices (e.g. USBs, keyboards, mouses

  • Also manages audio and disc drives

What are the popular metrics for measuring speed of computers?

• Increase clock rate

  • e.g. 1.87 GHz = 1.87 billion ticks per second

  • IMPORTANT: Computer being faster ≠ CPU being faster

  • Only the CPU gets faster - Unfair to compare clock rate

• MIPS (millions of instructions per second)

  • Better indication of speed BUT depends on which instructions are counted

  • Also unfair as there can be different results for different programs

• FLOPS (floating point operations per second)

  • Even better indication of speed “where it counts”

  • Also unfair

• None are ideal to measure other factors, e.g. I/O speed

What are limiting factors on speed?

Density limitations

• Number of transisters per square inch

• “Moore’s Law”: made in 1965, transistor number on a silicon doubles every 2 years

  • Still only a theory as it was only observed

  • Did not increase performance, only transistors

Power limitations

• 1/3 of power used to propogate clock signal around the processor

• Therefore power and heat problems increase as clock rates increase

  • Cooling becomes challenging to accommodate this problem

As performance demands increase:

• Inability to increase clock speed lead to the manufacturers making multi-core processors

• Can also increase performace by linking computers using high-speed networks

  • Led to “blade servers”

  • Application runs across cluster but some application cannot be easily decomposed this way