Original URL: http://www.theregister.co.uk/2003/09/15/gelsinger_100w_power_dissipation_ok/

Gelsinger: 100W power dissipation ‘OK’ for desktops

Managing expectations for Prescott

By Tony Smith

Posted in Channel, 15th September 2003 23:41 GMT

IDF Intel CTO Pat Gelsinger came as close as he's likely to get before the chip's launch to admitting that 'Prescott', the next generation of the Pentium desktop processor, dissipates more than 100W of power.

Speaking to reporters at Intel Developer Forum in San Jose today, Gelsinger didn't confirm Prescott's alleged 103W power rating; but speaking about processors in general, he said: "100W for a desktop is OK".

If that's not managing expectations for Prescott, we don't know what is. That Intel believes it's fine for a desktop CPU to dissipate 20-odd Watt more power than today's Pentium 4 does, is a sign that future chips - e.g. Prescott - will indeed pump out rather more heat than an 85W 'Northwood' P4.

Discussing this paradox - that a more compact process can yield a chip that runs far hotter than one fabbed using a bigger, 130nm process - Gelsinger said: "When you go through a process technology step, at the same [clock] frequency you can use a lower core voltage and as a result use less power. But you also get a bigger transistor budget - if I had twice as many transistors, I'd go and utilise those - and that takes the power up. I also tend to take the frequency higher, and that takes the power as well."

Leakage, too, is more of a problem with a smaller process, he added. "These very fast devices, at very small gate oxide [lengths], start to become very leaky. And we're seeing an increasing percentage of the overall power going into leakage power. So you're not getting the same sort of scaling gains as you have in the past."

Leakage is exponentially worse going from 90nm to 65nm than it was from 130nm to 90nm, he said. "Seeing the same sort of improvements in power that we've traditionally gotten for free, now becomes a hard engineering task."

In other words, don't expect a 90nm to run cooler than a 130nm part - expect the opposite.

Prescott's power envelope was revealed a little while back by Japanese web site PC Watch. To date, as Gelsinger noted, Intel has not made public details of what it expects Prescott's power dissipation to be. The P4 dissipates 85W, and is hot enough - at 103W, Prescott is going to need plenty more heat extraction technology than its predecessor.

However, Gelsinger said Intel has put in place a "very broad programme" targeting power reduction throughout the system. "We used to optimise for transistor cost, now we optimise for power".

So, Intel is working had to figure out how the rest of the computer can be made more energy efficient to cover the extra losses introduced by Prescott. Undoubtedly, the team behind the Pentium M and i855 Centrino chipset, both of which have made power conservation a key feature, will have much to contribute to this programme. ®