G. I. Meijer of IBM Research in Switzerland recently published an engineering paper called “Cooling Energy-Hungry Data Centers”, you can read the abstract here, and you can read the whole article if you’re a AAAS member.

He makes a great argument for liquid cooling in the data center, something I’ve been an advocate for a long time. It boils down (ha!) to something simple:

Heat capacity of air: 1 MJ m^–3 K^–1 Heat capacity of water: 4 MJ m^–3 K^–1

Clearly, the heat capacity of water, and other liquids, is far greater than air.

Interestingly, he’s not just talking about re-engineered cabinets which use liquid cooling to cool the air before and after it has been ducted through hardware, which certainly helps and while it isn’t common it isn’t rare, but rather he advocated the use of microfluidic heat sinks (using microchannel heat sinks and liquid cooling) at the transistor level to alleviate the heat issues caused by leakage currents at the gate oxide (we currently loose more in leakage currents than are consumed by computation). As we move from 45 nm through 32 nm to 22 nm packages in the next few years, this issue will accelerate - thermodynamics is still your daddy.

He estimates that using cooling water at 60C-70C will protect the microprocessors from overheating, and alleviate the need for chillers to operate at the extent they do today (or at all at certain times of year, probably depending on your local environments ambient temperature and humidity) - with a 50% reduction in data center energy consumption. A second benefit is that collection of the waste heat becomes easier, with applications in office and district heating and some industrial applications.

Anything we can do to reduce the 330 TW·h of energy in data centers globally (2009 estimate) is a good thing, right?

---

Science 16 April 2010: Vol. 328. no. 5976, pp. 318 - 319 DOI: 10.1126/science.1182769