Supercomputer eats more power than small city

Gelsinger co-efficient in action

A big thank-you to Adam Daniel who alerted us to The Register's guest appearance the New York Times, our second favourite NY paper.

This fleeting glimpse can be caught in a piece about trade-offs between heat and power in supercomputer design, courtesy of our ground-breaking article How to fry an egg using an Athlon XP1500+ . (Everything to do with heat, nothing to do with supercomputers. But you can see where the NYT is coming from.)

The way semiconductor design is going CPUs will generate more heat than a nuclear reactor by 2015. This calculation, a side-effect of the world-famous Moore's Law, is known, to us anyway, as the Gelsinger co-efficient. Chipmakers and their suppliers are developing new materials, shrinking the die size and investing in sundry manufacturing techniques to address this hot issue.

But this is not much help to some of the big tin in operation today. Where better to go than Los Alamos, home of the Nuclear Bomb and some absolutely supercomputers. One, a beast called Q, will consume enough energy to power 5,000 homes when it's fully up and running later this year, drawing 3 megawatts for the machine, and 2 megawatts for the cooling system. It lives in the Nicholas C. Metropolis Center for Modeling and Simulation, a three-storey 333,000 sqft structure which incorporates several cooling towers, and cost $93m to build.

On the drawing board is a 100-teraops machine. "To satisfy its needs, the Metropolis center can be upgraded to provide as much as 30 megawatts - enough to power a small city", the NYT reports.

Compare and contrast with a new Los Alamos supercomputer, the Transmeta-low power CPU- based Green Destiny. "Though Q will be almost 200 times as fast, it will cost 640 times as much - $215 million, compared with $335,000 for Green Destiny". And it doesn't need to live in a $93m temperature controlled, dust-controlled building. It measures two by three feet and stands six and a half feet high.

"There are two paths now for supercomputing," Wu-Chung Feng, the leader of the project told the NYT. "While technically feasible, following Moore's Law may be the wrong way to go with respect to reliability, efficiency of power use and efficiency of space. We're not saying this is a replacement for a machine like Q but that we need to look in this direction."

Here is the NYT in full. As always with the NYT, you have to register (hence the rather long précis) , but it's free.

And now for some Register articles about heat/power trade-offs in CPU design which you don't have to register for. There's less about supercomputers than NYT's sterling effort but more jokes to the column inch. ®

Transmeta blades power landmark supercomputer breakthrough
Intel touts Alpha, IBM designs to beat 'hotter than reactor'chips
Chip designers vow to cool overheating Gelsinger
British, American scientists discover Gelsinger co-efficient