Slashdot Mirror
Wriggling Heat Sinks
YourHero writes "Purdue researchers have come up with a new way to cool chips, in about 2 years. Just build a bunch of little piezoelectric fans (the waving kind, not the spinning kind). Since they don't spin, no bearings, less self-generated heat. Since they don't have magnets, no electromagnetic noise problems. And, of course, super-efficient. A press release and abstract for your reading pleasure. Formal presentation at THERMES 2002 Jan 15th."
Half seriously, though, you might think of superconducting chips to eliminate the heating due to the resistance in aluminium/copper wires. But AFAIK you can't build logic circuits entirely out of superconductors. The siliconductors (sic :-) we now use, require current to pass
through potential differences (energy gaps in the crystal structure). Power dissipated equals
current times potential difference, period. And there are lower limits for the voltage imposed
by the semiconductor used.
Until we get something entirely different, I'm quite happy to put my geekineering effort into the design of better cooling. I'm sure it can be almost as fun as inventing new kinds of logic chips.
Escher was the first MC and Giger invented the HR department.
I don't mind the noise, but dissipating heat in general would be a good thing.
The thing they need to do is make chips that run cooler. And yeah, Crusoe's do run cooler but they don't perform optimally in a task-switching environment.
Cooling the CPU is fine, but the heat has to go somewhere and a better solution is to go back to the drawing board and figure out how to reduce the heat output in the first place. PLEASE.
----
I don't have a solution, but I certainly admire the problem.
The only way I wont be surprised is if we dont see a repost of this tomorrow
I've been looking into this a lot recently, and there's some pretty (ahem) cool developments on the cpu front recently, with x86 architectures.
Some people point to the VIA C3-800, but if you have real computing needs, steer clear. It runs comparable to a Celeron 400, which is almost, but not quite adequate for general computing. Instead, check out the old reliable suppliers. The shift to .13u means a lot. Frequencies are so high and chips are so powerful that underclocking has become a real option. A good general target for fanless operation is about 12 watts. You can go higher with good case airflow, or lower if you're dealing with troublesome ambient temperatures.
Right now, you can take the Intel Tualatin pIII 1.13GHz (28W), cut the bus speed to around 100MHz, cut the voltage down to about 1.1v and be right in the target range. Of course you won't know exactly w/o experimentation on your cpu, but it *should* be doable. If you're worried about losing efficiency to bus speed, remember that you can compensate by running it on one of the PIII DDR chipsets that are now available (upping effective bus speeds to 200MHz) or waiting until February, when Intel says they'll release a similar part themselves. Additionally, the 512k (vs 256k) cache on the pIII-s will offset lower bus speeds. Just check out the specs of the PIII-M LV models at developer.intel.com and ask how they got to those low wattage numbers with the same core. Since the last fanless G4 was 400MHz and claimed (in its wildest fantasies) to be a supercomputer twice as fast as a pIII, a fanless 800MHz pIII is not insignificant.
Even better, surprise, is AMD. The current mobile palomino runs at 1.1GHz, 1.1v, 25w. This is clearly just an underclock of the current 1.75v desktop XPs. But what it tells you is that the AMD architecture is very open to undervoltage at lower clock speeds.
Now if you consider AMD's forthcoming die shrink, things really look good. Zdnet.de reported (unsourced) that the Athlon 1.73GHz processor would drop from about 75W to 45W after the changeover. Depending on how far you could drop the voltage, you could be looking at a 1-1.2GHz part running at about 10W! Fanless! Now imagine (a beo..no) 2 of these in a well ventilated case, with an MPX board -- 2GHz of dead silent AMD power! Wooo!
Alright, I'm calmed down. Back to your original point. It's really a shame about the alternative architectures. Every time I think of venturing into the embedded market, I get brushed off by the 2x price, 1/2x power rule. But since the ARM and PPC don't seem to be generating any economies of scale, at least mainstream processors are progressing fast enough to make cool, cheap and fast a real alternative.
Kill, Tux, kill!