15 April, 2008
The CPU is a "general purpose processor". It has instructions to run every program available for a computer. It can do word processing, display pictures, play music, or solve complex Physics problems. It has circuits that perform each type of operation, but obviously each type of program uses different processor capabilities. When you do word processing, the part of the CPU used to solve complex Physics problems goes unused.
Sorry, but real business problems don't require that much processing power. The Fortune 500 corporation in 1990 performed all its corporate data processing on a computer with less power than the cheapest desktop you can buy at Wall Mart today.
However, there are a small number of problems that can strain a home computer. For example, if you put a Blu-Ray disk into your home machine and try to view High Definition movies on your big computer screen, this would strain even the fastest CPU if it used the "general purpose" circuits. Fortunately, this type of processing tends to involve doing the exact same thing over and over again to a long sequence of data. This type of repetitive processing can be done much faster with specialized circuits.
In a restaurant, a “dishwasher” may be a guy who washes dishes or a machine where you stack the dishes and turn it on. The guy washes dishes one at a time, but he can do other jobs like cleaning the floor or taking out the garbage. The machine washes a hundred dishes at once “in parallel”, but that is all it can do.
In addition to the general purpose circuits, the modern CPU has other circuits that do a single operation on a block of data. These SSE or SIMD (Single Instruction, Multiple Data) circuits speed multimedia processing for sound and video decoding. While the CPU has a little of this SIMD capability, the Graphics Processing Unit (GPU) on your video card has massively more of this type of processing capability. For multimedia processing, the GPU can do 7 to 10 times as much processing as the general purpose CPU.
Moore's Law says that the number of transistors doubles every 24 months. Twice as many circuits could double the number of current cores, but Intel and AMD are too smart to just consider the obvious. The alternative is to keep two or four current generation (Core 2 or Phenom) complex cores, and then do something entirely different with the extra circuits.
One Intel experiment proposes to create a massive number of much simpler general purpose CPUs. Imagine 80 simpler Pentium-class processors that run at say 1/4 the speed of modern processors. If you had the software to keep them all busy, they could do the work of 20 modern cores with the same number of transistors that would be used to build only four additional Core 2-class processors.
AMD, on the other hand, proposes Fusion. They imagine using the extra transistors to bring GPU class specialized processing inside the CPU. At the low end, this produces a more powerful or less expensive entry level computer. At other price point it can significantly speed up the handful of applications that can use SIMD capability.
It is way too early to try and predict who has the right idea. There is time for a third alternative to emerge that will be better than both. The important idea is that after two or four of the current complex Core 2 class processors have been added to the chip, the next thing to add may be something entirely different instead of more of the same.