Desktop microprocessors enjoyed key advances in 2006. All but the low-end x86 chips migrated to dual-core, in which two processor cores are placed on the same piece of silicon -- in effect giving consumers two processors for the price of one. Both AMD and Intel have made clear that future advances in processor performance will come from the addition of more cores. True to its word, Intel shipped a four-core, or quad-core, chip in late 2006. AMD has announced it will ship a quad-core competitor early this year.
Consumers buying these chips, however, have been distinctly underwhelmed by the performance boost they received. Hard science backs up their disappointment. On multicore chips, each core runs slower than the 3.4GHz high-water mark for Pentium 4 and Athlon chips, to which we’d all become accustomed. The reason for this has to do with power consumption. If you were to look at a chart of power consumption vs. performance, you’d see a shallow incline at speeds below 3GHz, followed by a steep ascent beyond that speed. This “hockey stick” curve explains what’s going on. The vendors have taken two cores whose power consumption is below the turn and put them together on a single chip. Each core operates at well under 3GHz, but the two cores together exceed this performance level.
Here’s the rub: To enjoy this performance, desktop software must make use of the additional cores by using multiple threads or multithreading, where the program is broken up into discrete tasks (implemented as threads) that can execute on the various cores. Until then, single-threaded programs -- which include most desktop software -- run on a single, slower core.
On the server, in contrast, most software is already threaded, so the addition of new cores results in an immediate performance boost. For this reason, server processors, such as Intel Xeon, AMD Opteron, IBM Power, and Sun Sparc are the vanguard of multiple cores. For example, in late 2006, Sun announced that its forthcoming “Rock” chip would run 16 cores. It will be a while before desktop processors reach that level, but it will surely happen.
An intriguing prospect is the possibility of greater integration between general-purpose microprocessors, such as those running most computers today, with specialty graphics chips found on graphics cards. The latter processors are distinguished by exceptional capabilities in floating-point arithmetic that could serve as an adjunct to CPUs. AMD’s 2006 acquisition of ATI, one of the two leading makers of graphics chips, gave speculation about these possibilities a boost. A specialty hybrid of this kind was, in fact, unveiled in the Cell chip, which IBM, Toshiba, and Sony co-designed. It uses a PowerPC core supplemented with eight arithmetic coprocessors individually dedicated to multimedia tasks. The Cell made its large-scale debut in the Sony PlayStation 3, which was released late in 2006 and was widely acclaimed for its superior graphics.
Like 2005, this year demonstrated that the microprocessor market continues to be a locus of rapid innovation. The ascendancy of AMD in its perennial fight with Intel assures us that 2007 will continue the trend of delivering new and interesting features to customers at remarkably favorable prices.