University builds supercomputer using Raspberry Pi micro PCs

Shawn Knight

Posts: 15,289   +192
Staff member

A team of scientists at the University of Southampton led by Professor Simon Cox have built a low-cost supercomputer comprised solely of ARM-based Raspberry Pi micro computers. More specifically, the team has combined 64 units to create a parallel computer that has been racked and stacked using none other than Lego building blocks.

The system is called Iridis-Pi after the university’s full-size supercomputer by the same name (sans –Pi, of course). The entire operation cost just north of $4,000 to build but that includes 16GB SD cards for each of the 64 computers. Collectively, the supercomputer has 1TB of storage and 16GB of RAM – all connected using Ethernet switches which weren’t factored into the overall cost to build the system. The entire system runs off a single 13 Amp mains socket.

university raspberry legos raspberry pi supercomputer

Cox said his team built the necessary software for the supercomputer over the summer by starting with a standard Debian Wheezy system image. From there, they used a free version of the Message Passing Interface along with code written in Python using Microsoft Visual Studio to perform parallel computing tasks. Naturally, the first thing they did when finishing the supercomputer was to calculate the value of Pi, a well-known first test for any new supercomputer.

Cox has published a step-by-step guide on how to build your own Raspberry Pi supercomputer. Getting your hands on 64 Raspberry Pi computers, however, might be a bit more difficult considering how popular they are.

Permalink to story.

 
Love that, building a supercomputer stacked in lego building blocks...power and whimsy.
 
Great for students, team for ArGe and know-how about multi-nod development. And thats all.


About performance?? 64 x Arm11 Cores on par with ehhmm, no need to compare since we will be using Broadcom's Videocore 4 GPU's 24 Gflop general purpose computing power.

So under a %99 parallelized problem system will result 64 x 24 Gflop = 820 Gflop of computing power at best.

You can easily beat it with an i5 +7970 + OpenCL. Which will probably cost less and consume less power too.

Again, its great for team and educational purposes, main topic here is not about great efficieny. Everyone on team should have learnt much about distributed systems from node-building to managing them.
 
I kept waiting for a comparison between this super computer and a general mainstream computer.
A super computer is lots of other computers all linked in to spread the work load over more room therefore making it faster, think daisy chaining
 
A super computer is lots of other computers all linked in to spread the work load over more room therefore making it faster, think daisy chaining
That wasn't my question, I understand that. My question was about performance of the combined processing power of the 64 micro PC's compared to a standard off the shelf PC.
 
@kukrek, sounds like you were on the team. Fine answer to cliffordcooley's question. And it sounds like a terrific lab project for an intro or second course in parallel computing. At the end, issue grades or however you do it there, disassemble, throw all the Legos into one box, store the Raspberry Pi boards in another, and let the next class have at it. Great project!

I can see the future now. Every U in the world picking up on this, buying boards and Legos for their own projects, driving Lego stock through the roof... note to self - call broker...
 
Back