There seems to be hope left for faster processors with clockspeed far above 20GHz – upto 50GHz. The new technology is called STG – stacked transistor gate – a three dimensional design, an eventual successor to the standard planar transistor design. I found this article with a bird’s eye view from the STG design.
But even if this technology makes it into commercial products – it will be to late to us programmers dealing with the multi-core race. We will probably be fighting above 10 processors per die before we get such fast processors:
The company is touting the device as an eventual successor to the standard planar transistor design, a design whose well-known clockspeed scaling problems have put the brakes on the clockspeed race and have forced the entire computer industry into the parallel computing paradigm that programmers are still struggling with.
If the SGT is able to get the per-thread performance train rolling again, then this would shift some of the burden of providing overall software performance increases off of programmers and back onto process engineers. However, the SGT is an unspecified length of time away from commercialization, and, by the time it gets here, it’s possible that most programmers will be grappling with core counts well north of 10 cores per die (i.e., a Nehalem successor that has moved down into the mainstream). So whatever relief the SGT may eventually provide is almost too far off to matter in terms of slowing down the multicore revolution.
This is just another article pointig out what we are facing today. The multi-core dilemma has left the workspace of application server developers: most developers working on rich clients are deploying their products in similar hardware configuration.
It could be the reason why we – and other speakers and authors – are getting so much attention in talks about concurrency.
Steffen and I will be speaking again about java and concurrency on the upcoming OOP 2008 conference.