How you'll stop worrying and learn to love water cooling

It's been twelve years since IBM stopped using water cooling on mainframes, but Roger Schmidt, chief thermal architect and distinguished engineer at IBM, is certain that it's coming back. When I spoke with him for my recent column, The Liquid Data Center, he told me he that data centers will have no choice but to go back to the future. The choice, he says, is to continue on the path toward higher density and move to direct liquid cooling or start spreading out to bring per rack power densities down. The latter just won't happen, he says.

Liquid cooling lines have already crept into the data center rows to feed air conditioning systems built into server rack doors or heat exchangers mounted in-row or above high density server racks. Gradually, he says, we'll start seeing cold plates on processors and perhaps power supplies, and these will use liquid coolant to directly remove heat loads.

When it's time to cut over to direct liquid cooling of IT equipment, IBM hopes data centers will move to water, where its expertise lies, rather than liquid refrigerant. Liquid refrigerant is more expensive than water-based cooling, but it also evaporates quickly if it leaks. That makes it a little less scary than water for data center managers who weren't around mainframes from 1964 to 1995, when water cooling ruled. But, says Amory Lovins, chairman and chief scientist at Rocky Mountain Institute, Inc., "You can mix electricity and water if you engineer it correctly." No one on the planet knows how to do that better than IBM.

In the heyday of the water cooled mainframe in 1987, Schmidt says, a Datamation magazine survey showed that 92% of all mainframes used water cooling. That transition took time, however. "In 1964, when we broke into the market it took three or four years to educate people on the benefits and reliability of water cooling," he says. He expects a similarly long adoption curve for liquid cooling for today's distributed computing environments.

Water cooling went away when IBM moved from bipolar technology to more efficient CMOS technology. Such a fix won't work this time around, however.

Schmidt gives three reasons why data centers will want to consider liquid cooling again.

• Number one: "You get much better heat transfer at the processor level, much better cooling [so you can pump] much higher power into that processor. [You can support] a lot higher performance with liquid cooing."  Lovins agrees that the efficiency gain is dramatic. "Water can hold 3,467 times as much heat as air per unit volume," he says.
• Number two, according to Schmidt: "You take the heat directly to water. You don’t have to go to the intermediate step of transferring it to air and then to water. That helps greatly in the cooling efficiency of the data center. It eliminates the potential hot spots that it seems like a lot of customers are struggling with right now."
• "Number three is the energy efficiency you gain. When I eliminate the intermediate step of going to air and back to water I can eliminate some of the air conditioning units that use a lot of energy and go back to liquid cooling, which uses a lot smaller pumps. From a price/performance basis it will look attractive."

Are you ready?