As a child, I vividly remember being mesmerized by the design of Cray supercomputers. Most famous for their iconic designs, these machines looked like they’d been plucked straight from a Doctor Who set. The phrase “they don’t make ‘em like that anymore” certainly applies here.
Back in 1985 when the Cray 2 was first sold, these multi-million-dollar devices boasted cutting-edge specs such as an 80MHz clock speed and 2 gigabytes of RAM and could be found in the offices of NASA and the DoD.
If you’re reading this on an iPad then you might be assured (and probably not surprised) to know that what you’re holding in your hand far outperforms what was the fastest computer in the world 30 years ago. Another advantage the iPad holds is power usage. An iPad uses approximately 2 watts of electricity. The Cray 2 used approximately 200 kilowatts.
This power consumption, coupled with an extremely component-dense design, meant that air cooling would be impractical and led Seymour Cray to use liquid cooling in his design, proudly exhibited with the large “waterfall” heat exchanger that sat next to every Cray 2. For me, this only added to the (ahem) coolness factor.
So, it has been somewhat disappointing for me that over the last 30 years since the Cray 2, liquid cooling has failed to find mainstream usage in data centers. Is this situation about to change?
For anyone reading who isn’t familiar with the science behind liquid cooling, here’s a quick primer on why it matters. When we talk about liquid cooling, we are talking about the use of a liquid to transfer heat from something that is hot (such as a processor). Most cooling solutions use air, mainly because air is free and easy to obtain. Why would anyone want to use a liquid? It all comes down to specific heat capacity. Specific heat capacity (SHC) is a property of materials that tells us how much heat is needed to raise the temperature of a certain mass of a given material by 1°C.
As you can see, a higher specific heat capacity will result in a smaller temperature change for the same mass and energy transfer.
The specific heat capacity of water is more than four times that of air at room temperature and this is the crux of why Seymour Cray elected to use it in his pioneering design. You’d have to force a lot of air through these tiny gaps to keep everything cool.
So water is just better at carrying heat. Simple! Air is a terrible conductor of heat and has a low specific heat capacity compared with water. This is why your CPU can run at 65°C but you have to keep the room at 27°C—a bit silly isn’t it?
So why hasn’t liquid cooling taken off then? Here are a few reasons:
- Limited flexibility
- Most systems either are totally sealed, which makes it hard to quickly change a dodgy RAM stick, or are submerged in a futuristic but not very practical bath of mineral oil; submerged systems also must be accessed from the top.
- Direct systems are limited to solid-state components
- That is, no spinning hard-disk drives
- Most people can’t stomach the idea of liquid and computers together, even if it’s not an electrically conductive liquid. It’s just what we’ve been conditioned to!
These are all important points—even the last one—in an organization where there’s often at least one data-center “layman” involved in the decision-making process. So if air cooling “ain’t broke,” why fix it?
Well, the data center of the future might not look like the ones we have today. Water cooling is also finding much use in today’s high-performance-computing (HPC) applications, carrying the torch of the Cray 2. Despite advances in efficiency, today’s supercomputers still use an incredible amount of power in a relatively small physical space. The advantages of liquid cooling are still great in these applications.
Outside of supercomputers, there has long been the promise of high-density data centers that would also begin to push the boundaries of power density. Some doubts linger as to whether high-density data centers will ever become mainstream, but if and when they do, you can be sure that many of them will use water cooling.
Looking more speculatively, as IT becomes more virtualized, there may be less need for lots of different types of hardware and it will become much more practical to just buy standard building blocks and link them all together virtually in a private-cloud environment. If and when the industry moves in that direction, many of the practical obstacles to water cooling will fall away.
So, from a purely engineering perspective, liquid cooling is ideal. From a practical perspective, it might not be ideal for the majority of applications today. But there is hope on the horizon for water cooling, led by companies such as Iceotope.
About the Author
Ross Momtahan writes on behalf of DCIM supplier Geist Global.