Why MCP at all?
MCPs (Multi-Core Processors) are the latest buzz phrase in processor marketing, although in truth they have been around in various forms for several years. MCPs are processors which have several cores or processing units on a single chip. These replace the traditional Central Processing Unit (CPU). MCPs are processors which contain not one processor but several, some of which may be specially designed to perform certain operations very efficiently.
MCPs have made their way into many mainstream PCs. Dell are now taking orders for quad-processor-based servers. If you look at your domestic PC you will find that many have “multi-core” processing units. Although domestic PCs are shipping with dual cores, graphics processors frequently contain many more, in some cases tens of cores, particularly at the high end. A single Graphical Processor Unit (GPU) may contain many specialised processing units actually within it. Why all this complexity?
The answer comes down to two factors, complexity/performance and power consumption.
As application developers demand more and more compute power, CPUs have start to run out of steam. With growing clock speeds the power that is being consumed is escalating. Faster clock speeds require more power. This generates heat and the heat poses a number of engineering issues; for the processor itself and in relationship to other components and boards.. Faster speeds on the processor also require faster memory access times, which in turn means more power is required to the memory and memory is growing at a rate which is at least as fast as Moore’s law implies for processors. In the past few years we have gone from 166 MHz processors to ones clocked at around 3GHz, or about a twenty-fold increase, with an accompanying growth in power consumption and radiated heat.
Where many PCs are being used together or are clustered, power costs are starting to become a significant proportion of the cost of ownership. The heat has to be dissipated somehow, which means greater need for cooling. In an increasingly environmentally sensitive world, power consumption and heat dissipation on an ever-increasing scale is seen as unacceptable.
Moore's law, the rule that processor compute power doubles every two years or so, implies that power consumption will also rise. In fact it grows significantly faster than the compute power, although such things vary widely with loading, technology and so on. Processors used to run cool to the touch, now in a modest room the heat generated by a small system is noticeable.
Multi-core technologies make it is possible to build components that specialise in particular classes of frequently-used operations and to run those tasks extremely quickly. These can be built with using lower power, slower-clocked components. By using these as part of the processor, together with some judicious management, it is possible to create processors which meet demands for more compute performance while maintaining lower power consumption and dissipation than would otherwise be the case.