Supercomputing isn’t known for being environmentally friendly, but the Barcelona Supercomputer Center wants to change that.

The Barcelona Supercomputer Center is in the process of building a supercomputer powered entirely by cell phone processing chips. These are called ARM processors, as opposed to x86 processors or RISC processors that most supercomputers are comprised of. What’s the benefit of using ARM processors? The supercomputer will consume much less power, and will be on its way to becoming the greenest supercomputer ever built.


We blogged about Amazon’s supercomputer in the cloud a few months ago, and how the Top500 list monitors benchmarks and gathers a list of the fastest and most powerful supercomputers in operation around the world. The list for the 500 greenest computers in the world is posted on Green 500. The center is confident that its ARM processor supercomputer will be in the top ten on the list.

IBM’s Blue Gene supercomputer is currently number one on the Green 500 list. It’s just a prototype but last November the computer could do two quadrillion calculations per second (or 2 gigaflops) per watt. The Barcelona Center’s prototype is called Mont-Blanc and when all is up and running next month, the center anticipates the supercomputer completing closer to 7 gigaflops per watt.

The ARM processors specifically are Tegra 3 chips, which are designed to use as little battery power as possible. The chips Mont-Blanc is using use about 4 watts which is a miniscule amount compared to say the Intel Xeon chip (which many supercomputers use) which uses between 50 and 100 watts.

Another tool to improve the power consumption of supercomputers is to use GPUs instead of high-powered CPUs. Nvidia is a big time graphics chip and software company that recognizes that its customers may have other purposes for the GPUs they make that to run graphics on their PCs. The company has released a software development kit to help customers repurpose GPUs to act more like CPUs.

When all is said and done, Mont-Blanc creators expect the supercomputer to use between 2,000 and 4,000 processors. Most supercomputers use fewer processors but processors with greater processing power (and greater power consumption). Instead of cramming the Mont-Blanc full of low-power, low-performance processors, the center is using low-power, mid-performance processors to get the kind of computing power they want to see.

All of this is happening right before Nvidia releases its successor to the Tegra 3, the Cortex A15. It’s a 64-bit processor that is still built for ARM computing, but it still only consumes about 4 watts. The only roadblock to implementing the new chips when they are released is fine tuning the supercomputer software to make sure it all still cooperates with the new hardware.

The software in question, the purpose of the supercomputer in the first place, is used by university scientists for research. Specifically, it runs through chemical and physics simulations and problems.

If the supercomputer can really implement the Cortex A15 chips, it won’t just be on the Green 500 list, it would also be in range to break into the Top500 list, along with the big name supercomputers that don’t put as much emphasis on reducing power consumption. It seems like a far leap from the supercomputer’s current processing power, but the center hopes that if it were to really implement Nvidia’s new chips, the Mont-Blanc performance could be boosted 200 gigaflops per watt.

Although the prototype will be in operation next month, the center hopes to have a fully-fledged, deployable supercomputer in five years.

And what does ARM think about all the hype the company is getting from this supercomputer? The company’s main revenue stream is and will be for the foreseeable future, tablets and smartphones. ARM finds the project interesting, but not having an impact on the company’s plans or revenue streams.

If you’re interested in reading more about ARM processors that changing the hardware scene for more than just tablets and smartphones, check out the highlight we published on microservers.