Y Safle

You may not have known it, but Moore’s law appears to have reached its limits.

If you don’t know what Moore’s law is, it is a law proposed by Gordon Moore (a co-founder of Intel) that the number of transistors on an integrated circuit would double every two years. Double the number of transistors on a chip and you essentially double the speed by halving distance between transistors. This he predicted in 1965.

And he was right! What a star!!!

So essentially since 1965 we have seen chip speeds (including, of course, processor speeds) doubling every two years (or less). And now we have processors that run … well .. very fast.

There have been some problems on the way. The technique of etching a circuit into silicon is based essentially on a photographic technique, but at some point the transistors got so small that the wavelength of photographic light required was so tiny that it was in the X-ray spectrum. The result - the X-rays passed right through the silicon. But clever people solved these and other problems.

But now there is a new problem. Each transistor gives off a tiny bit of heat, and all those transistors working together are running the chips too hot. Heat dissipation is a major issue, and of late we have not seen the big increases in chip speeds - which is why we are going to parallel multi-core designs.

But wait a minute! Does an increase in the clock rate really increase the speed of the chip? Perhaps we could keep the clock speed static with more intelligent chip designs.

Why? Because of pipelining. One instruction is not executed each clock cycle any more. Instead, processor pipelines are created and in a clock cycle, one part of one instruction and another part of another instruction is executed.

Consider a launderette which opens with just one washing machine, spin tub thing (whatever that bit is called) and one dryer. I go in and wash my clothes in the first wash cycle. In the second wash cycle I use the spinner thing and Fred washes his clothes. In the third was cycle I dry my clothes, Fred spins, and Mary washes her clothes.

This is the way processor pipelines work.

And Apple made an excellent point about processor speeds and pipelines here:

Megahertz Myth

Their point - run the chips slower with shorter pipelines and you could improve performance.

But I guess the question is: if PowerPC architecture is so much better than the Intel arhitecture, why did Apple switch to Intel chips?

Hmmm.