- What is a CPU?
- Tracing an instruction
- L1/L2/L3 Cache
- Clock cycle speed
- Front side bus (FSB)
- The numbers game: Intel vs AMD
- Sockets and slots
- Dual-core and quad-core CPUs
- 64-bit processors
- Mobile Processors
Tracing an instruction
Here's how a CPU does its thing: the instruction pointer directs the instruction fetch to a spot in memory that holds an instruction. The fetch grabs the instruction and hands it to the instruction decoder, which examines the instruction and determines what steps are necessary to accomplish that instruction. (An instruction can comprise numerous steps that need to be accomplished in a specific order.)
The ALU then performs the work the instruction calls for: it adds, subtracts, or otherwise manipulates the data. After the CPU interprets and performs an instruction, the control unit tells the fetch to grab the next instruction in memory.
This process continues -- instruction after instruction, at a dizzying pace -- to create the results you see on your monitor.
Both of the major CPU manufacturers, AMD and Intel, have introduced proprietary technologies designed to optimise efficiency of CPUs at particular tasks and with particular software. It's not strictly necessary to understand where each and every bit that passes through the CPU goes in order to make an informed CPU buying decision, but it is important to understand the basics of each technological innovation in order to properly evaluate CPU performance.
To improve calculation performance, chip makers placed another arithmetic logic unit in the CPU. Theoretically, this meant twice as much processing could be done at once. Using multiple ALUs is like having two people lay a kitchen floor instead of one: the job gets finished more quickly.
In addition to multiple ALUs, Intel integrated the floating point unit (FPU) into the CPU. The FPU handles extremely large and extremely small numbers (those with many decimal places). While the FPU handles those types of calculations, the ALU is free to do something else at the same time, further enhancing performance.
AMD and Intel also sped up instruction processing by pipelining the instructions, or running them nearly parallel to each other. The execution of an instruction requires many separate steps -- fetching and decoding an instruction, for example. Originally the CPU had to complete an entire instruction before beginning the next one. Now, discrete circuits handle the separate steps. Once an instruction has moved from the first step to the second step, the transistors that performed the first step are free to start working on the next instruction, therefore speeding up the process. It's like climbing stairs: as soon as your feet leave a stair, the person behind you can use it.
Other additions that enhance performance include branch prediction (guessing which jump the program is likely to take); speculative execution, executing the predicted branch ahead of time; and out-of-order completion, the ability to complete a series of instructions out of a program's normal order.