The speed of an arithmetic unit depends largely on the quality and character of its and-gates and or-gates. A built-in "timekeeper" sees to it, just as the metronome does for someone learning to play the piano, that uniform times are kept. It measures the time a counting "beat" has to take. We have measured such beats previously - remember? First beat: The first two numbers are added ... and so on. In practice it usually happens that a counting operation is itself divided up into several beats: first beat - the pulses flow into the adder matrix. Second beat - they are added together. Third beat - the total is recoded. Fourth beat - they are stored. Or a similar procedure. But no succeeding beat must start until the timekeeper provides the signal, by giving the pulses a shove that means: "to work!" This is the only way of making sure that the counter will work smoothly and never lose its nerve, even during complicated operations in which several calculating operations are going on side by side.
If the timekeeper himself gets out of step, of course, he makes the whole computer useless.
The shorter the "beat" - the time interval - the more quickly the computer works. The more quickly it counts, the more valuable it is. The makers of computers therefore try to make their time intervals as short as possible.
The amount of time a fast computer would need to add up the 36 items of our grocery bill is therefore hardly worth mentioning. The addition of 36,000 numbers - mother's shopping bills in ten years - would take only parts of seconds. At this rate, of course, a punched tape is useless, for it needs at least 0.001 of a second to pass on one number. If only for that reason, its data are first accommodated in storages which can keep up with the speed of the arithmetic unit.
Adder matrices can also be saved with the aid of storages. A single matrix is enough for all digits from units to millions, provided that intermediate storages have been cleverly fitted in. If the matrix has added the units, the result is stored. The overflow too. Then the stored tens are fed into the matrix - the overflow from the units together with them - added and stored again. Finally the matrix adds up the hundreds, and so on. The automatic calculator then arrives at the total like a clerk who has no adding machine: first he totals the units, then the tens, and finally the hundreds.