What a CPU does

A 3 GHz processor ticks 3 billion times a second, one tick every third of a nanosecond.

Picture a computer class. (This is a made-up scene.) Sixteen numbered boxes are drawn on the board, each with eight bits written in it. Three students play the CPU. One holds a card marked PC, one a card marked IR, one a card marked ACC. The PC card says 0. The IR student walks to box 0 and copies what is there, 0001 1110, and the PC student changes the card to 1. The class looks up the first four bits in a table on the wall: 0001 means LOAD. The last four, 1110, are 14. So the ACC student goes to box 14 and copies the 7 she finds there. Then the whole thing starts again from box 1.

Box 1 says ADD 15, so ACC becomes 7 + 5 = 12. Box 2 says STORE 13, and the ACC student writes 12 into box 13. Box 3 says HALT, and everyone sits down. Twelve moves in all. None of the three needed to know that the program "adds 7 and 5". Each one followed one small rule. The boxes are the memory, the cards are registers and the table on the wall is the decoder. A real CPU works the same way, billions of times a second, and nobody inside it understands the program either.

The same short loop, billions of times a second

Gates can add, compare and choose, but something has to tell them what to do, in what order and with which numbers. That is the job of the central processing unit, the CPU. It repeats one short loop. The program counter (PC) holds the address of the next instruction. Fetch: the CPU reads the cell the PC points to into the instruction register (IR) and moves the PC on by one. Decode: control circuits look at the opcode bits and set up the right paths, for instance connecting the adder for ADD. Execute: the work is done. A number is loaded, added or stored, or the PC is changed so the program jumps somewhere else. Then the loop starts over. One instruction, JZ, jumps only when the accumulator is zero, and that single kind of jump is what lets a program decide and repeat.

The scene runs the same program the class acted out. Press Step and the PC, IR and ACC change exactly as the cards did: LOAD 14, then ADD 15, STORE 13 and HALT, four instructions of three phases each, twelve phases in all. The multiplication program uses the same few instructions, but its JZ and JUMP send the PC back to 0 twice before the counter in cell 15 reaches zero, so it runs 24 instructions in 72 phases. Six times the instructions of the addition.

The program sits in the same memory as the numbers

On the board, the instructions and the numbers 7 and 5 were in the same row of boxes. That is no accident. In June 1945 John von Neumann wrote the "First Draft of a Report on the EDVAC", describing a computer that keeps its instructions in the same memory as its data, and almost every computer since has worked that way. An instruction is just a number whose bits are split into fields. In this toy computer each cell holds 8 bits. The first four say what to do, the opcode, and the last four say which cell to use, the address. 0001 is LOAD, so 0001 1110 means "LOAD 14": copy the number in cell 14 into the accumulator, the CPU's working register.

How fast the loop goes

The steps keep time with a clock signal, a voltage that flips on and off at a steady rate. A 3 GHz clock ticks 3,000,000,000 times a second, so one tick lasts 1 ÷ 3,000,000,000 seconds, about 0.333 nanoseconds. Light covers only about 10 centimetres in that time. At one phase per tick, that is enough ticks to run the toy's 7 + 5 program 250 million times a second. Real processors don't finish one whole instruction per tick in the simple way our toy does. They overlap the phases of several instructions like an assembly line, which is called pipelining, and today's chips hold several complete processors, called cores, side by side. The toy has none of that, which is exactly why you can follow it by hand.