A very interesting press release from Sandia Labs last week discussed results from their research into how processor core counts correlate to performance. Briefly, their tests showed that solid performance gains (around 25-50%) are achieved in moving from dual to quad cores. But increasing core count from four to eight may add only a slight bump in speed, and the jump from eight to sixteen cores may actually result in lower overall performance. So, at least according to Sandia’s research, we are already close to the point where we will see diminishing returns from adding more cores. Intel has a six-core chip on the market now, as does Sun (albeit not an x86 processor), and IBM is rumored to be coming out with an eight-core chip at the end of this year. It looks like the traditional methods of jacking up chip performance (die shrinks and frequency bumps) will become the only methods in the short term. The problem with this is that we’re hitting serious constraints in terms of the amount of energy you can push through increasingly smaller pipes. Given this, how does the industry stay on the performance curve dictated by Moore’s Law? We believe it will be through a combination of advanced materials science and processor specialization. Better materials can handle the heat generated by high frequencies, and specialized parts can handle specific tasks far quicker. GPUs can handle compute-intense workloads much more quickly than general purpose processors. That same specialization will eventually make it into the broader computing market and allow the march of progress to continue – if not unabated, then only slightly abated.