Page 1 of 3
A quirky look at the power supply. Why do computers need power at all and how do power supplies work. The answers are surprising.
I want to tell you about a part of every computer that Babbage would have appreciated and even taken for granted. However the humble power supply turns out to be both a technological marvel and something that has deep philosophical implications as well as the ability to keep you warm on a cold day…
Why so hot?
When you think of a computer you don’t naturally think of huge pieces of machinery pumping out kilowatts of power but until quite recently computers were power hungry beasts. We all know that the first computers used valves and consumed the electricity needed for a small town to keep them all heated to a dull cherry red but why is power still an issue in today’s microprocessor age?
In fact, once you dump the idea of using heated valves, why does computing consume any power at all after all it’s not as if it did any useful mechanical work like lifting something!
To Babbage the whole situation would have seemed self evident. His first computer was a huge mechanical contraption with gearwheels and drive shafts. Obviously it needed something to drive it and in the first instance this was going to be a human turning a handle and in the fullness of time it would have been a steam engine.
In this case you can also clearly see why his computer needed power in practice. It needed the power to overcome the frictional losses in the machinery and here it becomes obvious that in theory at least his machine could have worked with a very small power input - but could it be zero?
To reduce the power needed to drive Babbage’s machine all we have to do is reduce the friction. In the same way in a modern computer we simply reduce the resistance of the connections – using superconductors if necessary.
In both cases however there remains the very interesting question of whether or not any power is needed for a computation once you have reduced friction to zero?
In other words does the act of computation when striped of all practical considerations actually need energy?
This is the deep theoretical and philosophical part of the issue that I mentioned earlier. The answer has been worked out but it isn’t as easy as you might think and it isn’t at all intuitive.
What matters is the concept of “reversibility”. The only sort of mechanisms that can work with small energy inputs that are so small that they can be reduced to zero in the theoretical limit have to be reversible and capable of being driven in both directions.
In the case of a computer reversibility comes down to not throwing any information away during the computation. The trouble with this criterion is that Boolean algebra always throws away information.
Consider a simple two input AND gate. If you know its output is a zero you can’t say if its inputs were 0,0 or 1,0 or 0,1 all you know is that it couldn’t have been 1,1. In this sense no Boolean machine is reversible and so computation always consumes power.
A few years back some logicians did invent a form of Boolean logic that was reversible. The basic idea was that you kept extra signals that allowed you to determine the state of the input to any gate you used.
A fun idea but not of any great practical value.
Now that we have considered the philosophy of power and computing what about there reality?
Even when gears, relays and valves were abandoned in favour of low power transistor circuits supplying power was a problem.
Early transistor computers may have been much more efficient but they still used a great deal of electrical power. They also had rather demanding standards for the quality of the power they consumed. Because of the nature of their circuitry they needed and still need a power supply that has a very constant voltage. Also because the voltage used by the logic circuits is low – usually 5V - a step down transformer is required and very high currents 30 to 40 Amps are often used. All of these considerations make building computer power supplies difficult.
The first generation of computer power supplies took a very basic approach to the problem. The simplest power supply uses a very big very heavy step down transformer, a bridge rectifier to convert the AC to DC and huge smoothing capacitors. In many mini and early micro computers the transformer alone was bigger than an entire modern PC power supply and weighed 10 or 15kgms.
The smoothing capacitors were typically tubes the size of a beer glass and they reached un-heard of values around one Farad (a typical capacitor weighs in at a millionth or less of a Farad!). You might think that statistics such as these are being quoted just to impress but no they had a very real influence on the way computers were used.
A minicomputer for instance needed a large air conditioned cabinet to keep its power supply happy and the first microcomputers had real problems in shrinking there power supplies to fit. Early machines such as the Altair or any of the Z80, S100 machines had power supplies that occupied half or more of the fairly large cases that were used.
Clearly something had to be done – power supplies had to get smart.