IBM Supercomputer Cooled With Hot Water 89
1sockchuck writes "IBM has deployed an innovative supercomputer cooled by hot water in a Zurich computer lab. The Aquasar supercomputer employs a chip-level liquid cooling system that can use water at temperatures as high as 60 degrees C (140 degrees F), and as a result consumes up to 40 percent less energy than a comparable system using room-level air-cooling. The system also uses waste heat to provide warmth to buildings, reducing Aquasar's carbon footprint even further."
Saw a presentation on this last year... (Score:5, Informative)
I was at LISA '09 [usenix.org], and Dr. Bruno Michel (works for IBM, mentioned in the article) made a presentation on this work (or at least very, very similar work). You can see the presentation, or download the MP3, here:
http://www.usenix.org/event/lisa09/tech/tech.html#michel [usenix.org]
Interesting talk, and well worth your time.
There is a video (Score:4, Informative)
first?
Re:There is a video (Score:3, Informative)
Re:There is a video (Score:3, Informative)
The temperature difference is higher, so the heat flow is more.
As for the frozen water: Hot water gets dissolved gas driven off and cold not. Cool both and the one with more solutes (the cold) gets frozen last.
Evidence of this: Hot tap that was not turned off at the mains snorts.
Hypothermia in hot water (Score:2, Informative)
Water is really an effective cooler even at what you might normally think of as quite high temps.
Reminds me that you can die of hypothermia even in tropical waters of 80 degrees if you are unfortunate enough to get trapped in such water for long enough.
Re:There is a video (Score:5, Informative)
Hot water doesn't freeze faster. However, water at 80'C will cool to 60'C much faster than water at 60'C will cool to 40'C, given standard atmospheric temperature and pressure for the ambient temperature of the room. The flow of heat from a hot medium to a cooler medium varies non-linearly with temperature. For example, as you approach the same temperature, the flow of heat approaches zero.
(In other words, if they piped through cold water which was heated to room temperature, a passive radiator would be useless.)
There is a drawback with hot water, though. The temperature gradient issue cuts both ways. As the temperature of the water approaches the temperature of the chips, the heat flow from the chips is reduced. Thus, water at 60'C will not draw off as much heat as water at 40'C, if the chips were to run at 80'C. You've got to balance this sort of approach fairly carefully.
I rather like the Cooling Tower approach (evaporative passive cooling). Basically, you blast the water through a nozzle that turns it into a fine mist. You collect the water that actually reaches the reservoir at the bottom and top it off. The drawback of this method is that it is somewhat bulkier than a radiator system. It is also not a closed system and therefore is a bit more expensive to run. On the other hand, evaporative cooling is much more effective than relying on simple heat flows, so you can get away with a lower temperature gradient at the cooling end. This, in turn, means you get a steeper temperature gradient for the chips, which means they're cooled much more effectively and can therefore be driven much harder without loss of reliability.
Re:There is a video (Score:1, Informative)
Hot water doesn't freeze faster
Who would've thought. Hot water actually does freeze faster:
http://www.phys.ncku.edu.tw/mirrors/physicsfaq/General/hot_water.html
Re:Saw a presentation on this last year... (Score:3, Informative)
Short answer...because they were engineered to that tolerance, and not beyond.
We could design chips that would run well at 80-100 degrees, but they would have to run slower and probably use larger transistors than current generations. The reason why all the chips run well at 60 degrees is because that's a reasonable temperature to be able to keep them at. They would run faster if we could keep them even cooler with practical approaches. You might have heard about "extreme overclockers" that use liquid nitrogen to get a desktop chip to run at 4-5 GHz. If it were practical and cost effective for everyone to cool their computers with liquid nitrogen, then they would all run faster, but you'd have someone on slashdot asking the questions "Why can't silicon/CMOS work reliably at room temperatures?".
because of (Score:1, Informative)
nonlinear, and even exponental change in resistance etc with increase in temperature
Re:Saw a presentation on this last year... (Score:1, Informative)
Higher temperature means higher electrical noise, slower transistor switching, and greater resistance leading to more power converting to heat, at a certain point this will become a positive feedback loop known as thermal runaway and the chip will convert itself into a pile of goo, like old AMD cpus were liable to do. This temp. is probably around 80+ C for most chips I guess...