Intel has produced a gem with the Nehalem microprocessor. While the architecture limits the chip to 8 processing cores today, Intel is doing research and advanced development with much bigger ideas. The samples of 48 core processors (see picture on left) that some people in the industry are envisioning capable of scaling to 100 plus cores one day is impressive. In the old days when I was doing large SMP server development, single core clock rate was a major vector that drove computational performance. Then, as predicted, a few years into the twenty first century the technology started to run into the laws of physics. Increasing the clock rate was no longer an available option. However scaling problems were now starting to be addressed by building multi core architectures. In other words placing multiple processing engines on a single die addressable by software. CPU to cache interconnects become smaller and parallel computing is beginning to approach a commodity. Introduce software to program these multi-core CPUs and parallel computing can realize significant gains.
An impressive fact of the Intel 48 core research chip is the power consumption of 125 Watts. That averages to about 2.6 Watts per core. This 48 core chip also has the ability to dynamically control voltage and frequency via software. So the power consumption can be much lower than 125 Watts. Keep in mind in the early 2000 time frame, 32 microprocessors in a SMP server would alone consume ~2 Kilowatts! In my opinion the rising costs of energy and cooling will become a barrier similar to that of the laws of physics. If your interested you can find more info on extreme core architecture here.
What are the industry folks in academics and research thinking 10, 20, 30 or more years out? The semiconductor world is quickly approaching process geometry limits. For example 45 nanometer technology is so small that you are manipulating layer thickness measured by atoms in the single digits. Margin of error keeps getting smaller and smaller. Is reliability a factor too? Absolutely. So where could technology lead us once we are manipulating single atoms to produce silicon? In research labs today the industry has already been able to manipulate single atoms.
One option would be to discover another variable beyond clock rate or multi core. Another option may have some promising hopes. Have you heard of Quantum Computing? All of us in Computer Science can relate to binary, octal, hexadecimal and decimal. Are you really ready to learn about the difference between bits and qubits? Click here.
Blog is also available at: http://blogs.sun.com/bobp