From Moore’s Law to the Law of Diminishing Returns

It looks like the good times are over and we all need to get used to a new reality of diminished expectations and slow, or even stagnant, growth.

You might think I’m talking about the economy, but I’m actually referring to processor technology.

Tech consultant Robin Harris hit the nail on the head earlier this week with his blog about the end of Moore’s Law, in which computing performance was supposed to double every three years due to increased transistor densities. But it should be clear to everyone by now that the 80s (and the 90s) are over and that even advancements like multicore technology, higher clock speeds and wider data paths will only produce minimal improvements going forward, forcing not just IT but the entire computing industry into a sort of holding pattern unless and until something really big breaks.

You can see this trend reflected in the latest product announcements. Fujitsu and Oracle, for example, have a new SPARC64 device slated for their new M-Series server that features a 3 GHz clock rate, a 12 MB cache and data optimizing technologies designed to benefit the Solaris platform and related applications. Even after all that, the companies tout only a 20 percent performance boost over current SPARC64 solutions.

Looking forward, Oracle is planning new SPARC T4 and T5 devices said to boost clock speed even further, deliver more cache and utilize a smaller manufacturing process, even as core counts shrink from 16 to eight. This offers a wider range of configuration options, and the company will no doubt try to squeeze as much juice out of them as possible. But a doubling of performance with each generation? Don’t count on it.

In fact, it seems that the IT industry at least has become less enamored with more power and is now focusing on less – power consumption, that is. Witness the increased attention being paid to ARM processors these days. Server manufacturers are jumping over themselves to outfit the latest designs like Marvell’s Armada XP, which clocks in at 1.6 GHz and delivers 16,600 DMIPS while staying within a 10-watt power envelope.

So it would seem that efficiency is replacing power as the new enterprise benchmark, which is fine so long as you don’t wind up deploying a dozen Armada machines when a single SPARC or Xeon will do.

Somewhere out there is a sweet spot that provides optimal data handling at the lowest cost. Unfortunately, that probably will be a different processor for each environment or even each application. And as data environments become more fluid through virtualization and cloud, it will be harder than ever to find the right match.