Supercomputer Built From Lego Blocks And Raspberry Pi ARM Boxes
    By News Staff | September 11th 2012 07:15 PM | 12 comments | Print | E-mail | Track Comments
    How cheaply can you build a supercomputer?  A group from the University of Southampton just made one using 64 Raspberry Pi ARM GNU/Linux boxes ($25 each) and Lego blocks. The machine, named "Iridis-Pi" after the University's Iridis supercomputer, runs off a single 13 Amp mains socket and uses MPI (Message Passing Interface) to communicate between nodes using Ethernet.

    The team was led by Professor Simon Cox and included Richard Boardman, Andy Everett, Steven Johnston, Gereon Kaiping, Neil O'Brien, Mark Scott and Oz Parchment.  Professor Cox's son, six-year-old James Cox, assisted with specialist support on Lego and system testing.

    The racking was built using Lego with a design developed by Cox and son.  The whole system cost under £2,500 (excluding switches) and has a total of 64 processors and 1Tb of memory (16Gb SD cards for each Raspberry Pi). Cox uses the plug-in 'Python Tools for Visual Studio' to develop code for the Raspberry Pi.

    Professor Simon Cox and technology specialist James with Iridis-Pi. Credit: University of Southampton.

    "As soon as we were able to source sufficient Raspberry Pi computers we wanted to see if it was possible to link them together into a supercomputer. We installed and built all of the necessary software on the Pi starting from a standard Debian Wheezy system image and we have published a guide so you can build your own supercomputer," said the elder Cox. "The first test we ran - well obviously we calculated Pi on the Raspberry Pi using MPI, which is a well-known first test for any new supercomputer. The team wants to see this low-cost system as a starting point to inspire and enable students to apply high-performance computing and data handling to tackle complex engineering and scientific challenges as part of our on-going outreach activities."

    Young James added, "The Raspberry Pi is great fun and it is amazing that I can hold it in my hand and write computer programs or play games on it."

    He's not kidding. The six-year-old used Python and Scratch over the summer to program the Raspberry Pi himself. 

    Have some Legos and $1,600 laying around (hey, the UK charges more for everything but they get free check-ups)? They put the instructions up here and now you can build your own supercomputer. Take that, Instructables.


    That is so cool!
    Never is a long time.
    Frank Parks
    Awesome indeed.  Take that IBM.  In your face Cray.  Wanna play Fujitsu?
    Though, they don't really list any performance spec's, and I'm guessing a single node isn't all that powerful, so I'm not sure Cray has anything to worry about just yet....
    Never is a long time.
    Right, it is just being parallel and at a tiny cost that makes it cool.  Oh, and Legos.
    No disagreement there, the 64 plugin power supplies powering a super computer is pretty cool too (and the first thing I'd fix :) )

    It might make a great system to run GCM's on as well.
    Never is a long time.
    Frank Parks
    Ya.  No mention of how many pflops or of how long they let it run on the pi problem.  My comment was just a tongue in cheek shot at the biggies in the supercomputer business.  In fact, Cray probably doesn't have to worry about the next several competitors either.
    But the truth is that the computer industry is full of stories(and dead bodies) of 'toys' growing into giant killers.
    WinTel decimated the workstation industry, the Nintindo 64 is a $200 version of an Iris workstation, and I was surprised to read an iPhone 4S has about 4x the floating point of the Cray1, and it runs on batteries.
    Never is a long time.
    Frank Parks
    I didn't know that about the iPhone but it doesn't surprise me.

    I agree with the other parts.  Right here in my home town is the processing center for an international oil and gas exploration company.  About eight years ago my brother, who worked in the systems area, told me of an experiment they were working on.  One of their brighter analysts suggested that all of the power and space that they had dedicated to one Cray and two IBM's was a waste of resources.  One year later they sold both IBM systems and were keeping the Cray as a backup for the (approximately) 400 Dell desktop systems that were processing all of the geophysical data for their worldwide operations.

    Cray and IBM did indeed see the desktop computers as 'toys'.  Properly configured and networked, the toys out performed their huge brothers.  I think there are many more examples out there too.
    I remember the Workstation vendors putting PC's down. Then they started OSF to create a common platform to develop code on because they knew the tsunami was coming, and they still couldn't stop their bickering, and then they were pretty much all gone.

    I did some consulting in Chippewa Falls, got to see them building part of Redstorm, they finally at least for a while paid attention to the lessons of history.

    And then there's those clever guys in the Air Force with their PS3's.

    Never is a long time.
    Trivia: the bulk of the Redstorm system was designed by one guy, a Tera engineer (that is why the HQ is in Seattle - they bought that part from SGI and changed their name to Cray) and he used our stuff from Ansoft. Who was the person who had the contract from Cray to make the analysis happen?  Yours truly. But we called it Thor's Hammer.  Fitting, because I make everything about Thor whenever possible. And with a name like Hank, 'hammerin' gets thrown in front of it a lot.
    Cool, I saw them working on one of the upgrades, a gymnasium full of rack mounted nodes.
    Never is a long time.
    Gerhard Adam
    Despite what it sounds like, having a Cray and two IBM systems is small by most system standards [without knowing more about the specific models involved].   Similarly 400 Dell desktop systems is also a small configuration.

    Unfortunately, the problem most often faced is bandwidth and not computing power.  So, if there are large amounts of data, having 400 desktop systems would be a catastrophe.  In fact, one can't even reasonably get much parallel processing [of the same task; i.e. database serialization] on such a multitude of systems because the inter-system communications would eat up all your horsepower.

    Even though there are perpetual claims made about this type of system environment, the reality is that the large systems and supercomputing environments continue to grow and do well. 
    One of their brighter analysts suggested that all of the power and space that they had dedicated to one Cray and two IBM's was a waste of resources.
    That's also a problem, because often an individual or groups of individuals can develop a comprehensive solution which becomes unsupportable later [i.e. when they are no longer around].
    Cray and IBM did indeed see the desktop computers as 'toys'.
    In many respects they are, because they don't adhere to the same standards and requirements that large systems do.  There's no question that these smaller systems can wield a great deal of power and solve many problems, but when it comes to major computing efforts, they're simply not ready for prime time.
    Properly configured and networked, the toys out performed their huge brothers.
    I'll believe that when I see it.  Keep in mind that a fully configured IBM mainframe can also be "networked" in a system complex [sysplex] which provides up to 1024 processing engines under the control of 32 fully-configured operating systems.  With connections of up to 64K worth of external devices and 4 TB of memory each, these are formidable configurations.

    NOTE:  These systems fully communicate with data sharing and messages, in addition to being able to manage each other with respect to "joining", "leaving", or "failing" in the configuration.  In addition, such systems don't have to be in the same facility and can be separated by over 25 miles.
    Mundus vult decipi