This past Friday the Leibniz Rechenzentrum (LRZ), a large scientific computing center serving the universities across Europe, inaugurated their latest acquisition — SuperMUC, a 150k core supercomputer, currently the world’s 4th fastest machine, #1 in Europe. Photos of the event can be found here.
SuperMUC Addresses Energy Challenges
Why am I proud of this? Well, SuperMUC is not only a fast machine, but it reaches the lowest datacenter-level carbon footprint, making it extremely energy efficient. It uses 40% less energy than an equivalent air cooled machine because SuperMUC is hot-water cooled computer. The idea to cool computers with -water is not new and in fact, IBM has been researching the concept since the 1960s – see timeline.
Commercial data centers already account for more than 2% of the electrical energy use in the United States alone. About half of that energy is spent to power the computers, the other half is spent on air-conditioning to remove the heat that gets generated when the computers run.
Our idea was to avoid the air-conditioning as much as possible and instead of using conventional air cooling we cool each major chip on the printed circuit board with water. So we pipe warm water anywhere from 35 to 60 degrees C through the computer, where it gets heated by 3-10 C. When you have water with 40 to 70 C, you can easily use it to heat buildings or run it through radiators on top of the data center where it gets cooled down by the ambient temperature, to be used again to cool the computer.
In hot geographies or in a data center that still runs legacy hardware you can use the hot water in a so-called adsorption chiller to generate cold air for air-conditioning, again avoiding the use of much additional electricity.
So while I am proud of SuperMUC, I am equally proud that we managed to bring this technology from idea to product in a only five years. Many of you may think this is a long time, considering how many new things our industry puts out every year, however be assured, all of these things, when they go beyond pure software innovations, have been in the pipeline for many, many years.
We came up with our own concept of chip-level water cooling in 2005 and build our first hot-water cooled zero-emission datacenter prototypes in the lab in 2006 and 2007.
A few months later ETH Zurich, the renown Swiss university, showed interest in the concept and we created Aquasar, which went live in early 2010. ETH was a great partner to help scientifically validate the concept in real life.
At the end of 2010 LRZ ordered SuperMUC and only 18 months later the system came to life and made it onto the top 500 supercomputer list. This spectacular achievement was a result of a concerted world wide collaboration between IBM’s Research labs, business units, and country organizations.
3D Chip Stacking
Now the research team is onto even more interesting concepts to packaging the next generation of computer chips.
They will consist of multiple layers of silicon in a single package. The layers will be cooled by running the coolant through the package. As supplying electricity into modern CPUs is already a challenge, we will explore a dual use of the coolant, it will also chemically supply power into the chip where it then gets converted into electricity. Similar to what happens in the human body, where our blood supplies the organs with energy and is also responsible to remove heat. Read more about it in this paper.
I wish we would see this idea in products again in 5 years, however the scientists tell me a lot more basic research is needed so it will take a little longer this time. The vision is a great one though, we want to build chips that ideally will be as energy efficient and dense as the human brain that uses only about 20 Watts instead of modern supercomputers that use a ten thousand times more power.
SuperMUC so is just the start, but one I am extremely proud of.