A lesson in making one's point thoroughly.

[brooksmoses]

So, I was looking through an IBM presentation on the future of high-performance supercomputing, and came across this slide that explains why they think a new approach is needed in dealing with computer-chip power consumption.

It illustrated the point with the following graph, of the power density (power usage per unit area) of current and past computer chips, with an admittedly somewhat generous extrapolation:



Somehow, I don't think we'll actually be meeting that extrapolation, rather than continuing on what they claim is the current technology path! (It's not the only extrapolation we won't be meeting; see this article from The Register on another graph in the presentation that compares Itanium sales to historical predictions.)

The worst thing is that, unfortunately, we can't use the heat for baking cookies with, either. Chips start to die at about 180F, but cookies need at least a 275F oven, and -- as Flanders and Swann point out, (cue music) Heat won't pass from a cooler to a hotter (you can try it if you'd like, but you'd far better notter)!

Comments

[oldsma.livejournal.com]

I think that a water-cooled PC would be interesting. What do you think they will pursue instead of heading out toward the rocket nozzle?

MAO

[brooksmoses]

Water-cooling would be interesting, if done on a mass-market level. (At least one of the very small PC cases comes with a water cooler because there's no room for the airflow, so the possibility is certainly there.) I suspect the chip makers are unlikely to release a chip that requires water cooling, though, even if it does become more popular in low-profile case designs. But I could be wrong; it's not necessarily that much different from requiring heat-sink fans.

In the presentation, IBM went on to talk about some of the power-reduction strategies that they have in their upcoming PowerPC chips; it looks like there's a lot of potential in lowering leakage currents and such, so that the power per transistor drops significantly. Also, they said they're investigating things where only the time-critical threads are run as fast as possible, and the low-priority threads are run in a lower-power mode.

I'd predict that there's a surge in research in those sorts of techniques for reducing power draw, and that we'll get a drop in power density such as the one for the 386 shown on the chart, and then things will gradually creep up as the limits of that technology start getting pushed again, and in ten years or so people will start making dire "if we keep doing this, we'll be up at rocket-nozzle power densities" predictions again.

And there's also the possibility of something strange coming out of the proverbial left field; a new chip material that can run at significantly hotter temperatures (and is thus much easier to pull heat out of with air-cooling), or if someone manages to design a practical 3D chip -- as opposed to the current layered 2D approach -- which will necessarily need built-in liquid cooling channels, probably with something a lot fancier than water in them.

[oldsma.livejournal.com]

It still sounds incremental, until you get to the 3D idea. It will be interesting.

Sometimes I think about what has happened to technology in my conscious lifetime. I'm glad I've gotten to see it.

(And yet it was billing strategy enabled by software advances that made cell phones pop up on every belt; that is what paid for the development that pushed the size down and the battery life up.)

MAO

[julesjones]

Snicker. Wonder if someone's been taking lessons from Huff's "How to lie with statistics"?

[bdot.livejournal.com]

OT: i need an e-mail address for you.... magazine order!!!!!
thanks!
bdot