Chips By Jun. 4, 2014 11:30 am
Twitter engineer crams 8-node Parallella supercomputer into a PVC tube | Chips | Geek.com

Building a supercomputer in your house wasn’t really a thing until a few years ago, but thanks to inexpensive ARM-based computers like the Raspberry Pi and Parallella it’s no longer the stuff of sci-fi dreams. In fact, it’s pretty easy to put one together yourself for a couple thousand bucks, and that’s exactly what Twitter engineer Brian Guarraci decided to do.

He calls his creation Parallac, and it’s an homage to supercomputers like the Cray-1 and Connection Machine. At its core are eight Parallella single board computers. They’re complemented by a pair of Intel Core i3 NUCs, which Guarraci added to provide additional RAM and storage for the Parallella nodes — which are fairly limited on both those fronts.

Twitter engineer crams 8-node Parallella supercomputer into a PVC tube | Chips | Geek.comEven under a full processing load, Parallac doesn’t draw a heck of a lot of power. Each Parallella board pulls about 8 watts, and the NUCs each add another 27 watts or so. 118 watts total isn’t too shabby, considering that covers a total of 130 processing cores.

For Parallac’s housing, Guarraci turned to a material that’s easy to find at your local hardware store or building center: PVC pipe. Typically it’s used for sewers and drains, but it also works nicely as a DIY computer enclosure. A 3-inch tall section of 12-inch pipe forms the bottom of the unit, and Guarraci routed a channel into it to accommodate the base plate that the cluster itself is mounted on. It also houses the two NUCs and their power bricks.

The cluster itself slides into a 13-inch-long section of 6-inch pipe. The Parallela boards are mounted on to copper bus bars, and those are affixed to a 16-port Netgear Gigabit switch with strips of velcro. Gaurraci still has some finishing touches to add, like wiring in one Adafruit Neopixel LED strip for each Parallela.

So how does the system perform? Guarraci says that Parallac has achieved around 208 GFLOPs. To put that in perspective, it’s roughly 200 times more than the first Cray-2 supercomputer could manage back in 1985. Parallac may not be the fastest supercomputer in the world today, but it’s an amazing example of just how far we’ve come in a relatively short time.

Soon you’ll be able to build your very own Parallac. Guarraci plans to open source everything, and he’ll be posting it all on his project website (it’s just a simple ‘hello world’ at the moment).

Twitter engineer crams 8-node Parallella supercomputer into a PVC tube | Chips | Geek.com

发表评论

电子邮件地址不会被公开。 必填项已用*标注