Architecture & Design
It's a common fallacy that each processor model is manufactured on separate production lines. The fact of the matter is, all processors of the same family are not only exactly the same in design, they are also fabricated on the same silicon platter. For example, there is no physical or electrical difference between an Intel Core 2 Duo E6600 and the Intel Core 2 Duo E6700 processor - just a difference in clock speed. And how exactly did they come by those clock speeds?
Frankly, processor speed grades are the byproducts of engineering and marketing. Processors of different speed grades actually start from the same platter. After they are fabricated, the chips are tested. Damaged chips are discarded or sometimes salvaged and sold as lower performing parts (for example, with half the L2 cache disabled).
Working chips are then speed-binned according to the clock speeds they are capable of running. After all, why sell all these chips at the same price when you can give certain chips a better speed rating and sell them at a much higher price?
However, that is not always the case. This is where market forces come into play. Sometimes, demand for a particular speed grade is much higher than what the yield of suitable chips can provide. Chip manufacturers will have to reclassify faster chips at lower speed bins to make up for the increased demand. This is one of the reasons why certain processors are more overclockable than others.
Manufacturers also have a habit of picking out especially good chips and sending them out as review samples or for marketing projects that are designed to showcase best-case performance. Sometimes, these chips are labelled as engineering samples. That's why many overclockers love looking for such processors.
However, this habit of misnaming specially-picked processors as engineering samples, confuses them with real engineering samples which are chips randomly picked for stress-testing and other tests. These true engineering samples are often damaged in one way or another. So, don't be so eager to pick up on that special offer to purchase a rare engineering sample without first reading our article on Core 2 Duo engineering samples.
A bus is a term for electrical interconnects that allows a processor to communicate with other components like the motherboard chipset. The faster the bus speed, the better the performance, although the degree of performance improvement does not necessarily scale linearly with the increase in clock speed. I'll give you an example.
Memory bandwidth has always been a key determinant of a PC's performance. Traditionally, the memory controller is part of the north bridge chip of the motherboard chipset. The CPU communicates with it through the system bus, also known as the front side bus. Because memory accesses are the major use of the system bus, overclocking the front side bus has always been a great way of increasing overall system performance.
On the AMD64 platform however, the processor has an integrated memory controller that communicates with the system memory via its own dedicated memory bus. This ensures a high degree of memory performance. However, this also means memory accesses are no longer hogging the front side bus. Therefore, increasing the system bus speed no longer brings much performance benefit.
On the other hand, the Intel platform still retains the memory controller on the north bridge chip. Hence, overclocking the front side bus with this platform will still deliver a significant performance boost, certainly much more than what you will see with the AMD64 platform.
This metric is usually measured in megahertz (MHz) or megabytes per second (MB/s). The latter is a better indication because it directly tells you what you want to know - the performance of the bus. The former isn't so clear-cut because the bandwidth of any bus is not only determined by its clock speed, but also the width of the bus. For example, AMD's HyperTransport bus has a 16-bit width, while Intel chipsets use a 64-bit wide front side bus.
<<< Architecture & Design - Pipeline Length, Production Process : Previous Page | Next Page : Architecture & Design - Cache, Overclockability >>>