Oct. 27, 2011 Today's multicore processors are not being utilized in a sufficiently intelligent way. They get too hot and run slowly because they are used inefficiently. At the same time, transistors are becoming so small that they will ultimately become unreliable. Major European research organizations are now attempting to create a revolution in computer architecture.
Are you disappointed with the performance of your new computer or phone? Then perhaps rightly so. Their clock frequency has not risen significantly since 2000. Overheating problems and limited battery capacity are forcing manufacturers to limit power usage substantially, typically to 100 W per chip.
"You can't sell a mobile device with bulky sophisticated cooling. So we can't increase the clock frequency, a problem that multicore processors were designed to avoid," says Per Stenström, Professor of Computer Engineering at Chalmers University of Technology and project coordinator for a new center called Eurecca.
"So far, multicore systems have merely yielded marginal performance improvements. It will be increasingly difficult to keep all the processors busy without exceeding the power limit."
Today modern microprocessors typically have six cores. The problem is that no one can as yet optimally program even two-core (two parallel processors on the same chip) computers. The same problem has also baffled giants such as Intel and Microsoft. Basically what is needed is moving from sequential to parallel programming -- i.e. the art of getting several processors to operate together as efficient, well trained sports teams instead of inefficiently coordinated superstars. This second technological challenge goes by the name of concurrency.
"The industry is not yet mature for this step, and we lag behind in terms of education and research," says Per Stenström.
That's why three leading universities in computer architecture are now founding the European Research Center on Computer Architecture (EuReCCA). The planned center so far comprises fifty research scientists, but computer technology institutions from all parts of Europe are joining at a rapid rate, all coordinated by Chalmers. The aim is to make pioneering progress in the development of systems involving multicore processors and paradigms for their productive programming, which should ultimately result in new generations of products and innovative company start-ups.
Eurecca should above all boost the competitive power of the European computer industry, which is a world leader in computers for embedded applications such as cars and mobile phones. A modern car contains no less than about fifty processors. British ARM processors are found in mobile phones from manufacturers such as Nokia, SonyEricsson and Apple. European research projects normally run for three years, but Eurecca will be set up as a longer term Eurolab structure.
"We want to have a permanent structure in order to sustain a long-term focus on bringing innovations out to industry. This has not worked in a satisfactory way in Europe so far. To build up functional systems of knowledge transfer is not something that can be done in three years," says Per Stenström.
The Eurolab structure is also needed to build up streamlined education in computer technology at master and doctoral level throughout Europe. The aim is to get together to nurture a new generation of computer architects ready to lead system innovations based on the parallel philosophy in high demand.
The third technical challenge is miniaturization. Transistor dimensions are halved every 18 months. Soon, transistors will be only a few tens of atoms in size. Along the way they become increasingly unreliable, which will present an enormous challenge for computer system designers in the future. No one currently knows how to tackle this.
"We quite simply have to create a revolution in computer architecture," says Per Stenström, who expects national and European funding for Eurecca, as well as investments from industry.
Other social bookmarking and sharing tools:
Note: If no author is given, the source is cited instead.