Factors affecting processor performance

Overview

Recall questions on cores, cache, clock speed, word length and bus widths, and on why each of them stops delivering gains at some point.

What you need to know

Answer each one before you reveal it.

Questions from this subtopic

Each one with the model answer underneath.

1. 1

   Name the six factors affecting processor performance.

   Model answer

   The number of cores, the amount of cache memory, the clock speed, the word length, the width of the address bus and the width of the data bus.

2. 2

   Explain the effect of multiple cores on processor performance, and its limits.

   Model answer

   A core is a complete processing unit with its own arithmetic logic unit, control unit and registers, so a multi-core processor can genuinely fetch, decode and execute several instruction streams at the same time rather than merely appearing to by rapid switching. Four cores can in principle do four times the work per second. The limits are severe in practice: the gain only materialises if the software has been written to divide its work into threads that can run independently, and many tasks contain sections that are inherently sequential because each step needs the result of the previous one. The cores also share main memory and the buses, so they contend for access, and there is an overhead in coordinating them. Doubling the cores therefore rarely halves the running time.

3. 3

   Explain the effect of cache memory on processor performance, and its limits.

   Model answer

   Cache is a small amount of very fast memory sited on or very close to the processor, holding copies of the instructions and data most recently or most frequently used. When the processor needs something it looks in the cache first, and a hit is served in a fraction of the time a main memory access would take, so the processor spends less of its time waiting. More cache means more of the working set fits, so more hits. The limits are that the benefit depends entirely on the hit rate, which depends on the program showing locality of reference; a program that jumps unpredictably over a large data set gains little. Cache is expensive per byte and consumes space on the chip, so it must stay small, and there are diminishing returns as it grows because the additional data held is used less and less often. Every miss still costs a full main memory access, plus the time spent looking in the cache.

4. 4

   Explain the effect of clock speed on processor performance, and its limits.

   Model answer

   The clock produces the pulses that every operation is synchronised to, and each stage of the Fetch-Execute cycle takes a whole number of clock cycles. Raising the clock frequency, measured in gigahertz, means more cycles per second and therefore more instructions executed per second on the same processor. The limits are that heat increases sharply with frequency, so cooling sets a practical ceiling and the chip may throttle itself; the physical time it takes signals to settle sets a hard limit; and, crucially, clock speed cannot be compared between different designs, because different processors need different numbers of cycles to complete an instruction, so a 3 GHz processor of one design may be slower than a 2 GHz processor of another. Beyond a point the processor is limited by how fast memory can supply it rather than by its own clock.

5. 5

   Explain the effect of word length on processor performance, and its limits.

   Model answer

   The word length is the number of bits the processor handles as a single unit: the width of its registers and of the arithmetic logic unit. A 64-bit processor operates on 64 bits in one instruction where a 32-bit processor needs several instructions to do the same job, and it can hold larger numbers and larger addresses in a single register without splitting them. Longer words therefore mean more work done per instruction. The limits are that longer words require wider registers, wider internal paths and more complex circuitry, which cost more and use more power; the gain is only realised on data that is actually that large, so a program working with single characters gains nothing; and programs and data structures may take up more memory because every value occupies a full, wider word.

6. 6

   Explain the effect of address bus width on processor performance, and its limits.

   Model answer

   Each line of the address bus doubles the number of distinct addresses available, since n lines give 2 to the power n locations, so a wider address bus lets the machine address more main memory. That improves performance indirectly but substantially: with more memory addressable, more of a program and its data can be held in main memory at once, so there is far less swapping of pages out to the much slower secondary storage. The limits are that the width helps only if that much physical memory is actually installed, and that the address bus carries no data, so widening it does nothing to speed up an individual transfer; a machine that is not short of memory gains nothing at all from a wider address bus.

Other ways to work this

The same content, four other ways.

What is inside Levo

Six ways to work the same material.