Southampton engineers build a Raspberry Pi supercomputer
September 12, 2012
Computational Engineers at the University of Southampton have built a supercomputer from 64 Raspberry Pi computers and Lego.
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 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). Professor Cox uses the free plug-in ‘Python Tools for Visual Studio’ to develop code for the Raspberry Pi.
“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,” says prof. James Cox.
If you want to build a Raspberry Pi Supercomputer yourself, see here.

Comments (14)
by Graham Rounce
How many flops??
by Arcadium
Did a calculation on it recently to answer that very question. Note these are optimistic (peak theoretical) estimates.
1x RPi (Rasberry Pi) CPU: 175MFlops GPU: 26GFlops for 1.5W
64x RPi total floppage: (26+0.175)*64 = 1675 GFlops for 96Watts
giving an efficiency rating of 17.45 GFlops/W
in comparison an AMD 7970 (GHz edition) has slightly more than 4TFlops (4000GFlops) for 225 W
giving an efficiency rating of 17.78 GFlops/W
So not bad in terms of overall system performance, but it is still about 2x more expensive per flop vs an equivalent desktop based supercomputer.
by asiwel
Of course power consumption is a really big deal at server farms, in my car, etc. But I guess I am sitting here with a 250 w light bulb in the lamp by the computer I am using and not thinking much about the power being consumed by either one. I am concerned about how “slow” my computer is sometimes and my own wasted time sitting here watching it crunch data. One of those 225 W desktop supercomputers would be a nice thing to have about now.
by melajara
The bright side is that, indeed, “supercomputing” (please define) has never been so affordable.
On the other hand, the dark side is to see how devastated is now what used to be the workstation market (which itself depleted the minis who feeded on the mainframes).
What this article is telling me, is that it is so depleted a market that it has to be addressed by academics engaging their students in hobbyist activities like they did in the pre PC era with the Acorn!
This is just to say that we are not really ramping up in Mips to Exaflops as quickly as possible, since the main (i.e. most competitive and innovative) market is centered on a sliding downsizing window. Today’s real competition is not really on CPU anymore but on SOC for smartphones and tablets and the rest of the market is kind of fossilized, alas, except, maybe for the transversal GPU market.
This is one bleak perspective Kurzweil didn’t foretell with his $1000 PC computing power metric he consistently used from year 2000 to the putative singularity.
The main reason for this phenomenon is the capitalist intrinsic logic. Once a market segment has been thoroughly addressed, soon enough it becomes saturated, hence the pressure to open new outlets for cheaper and cheaper devices (a quasi automatic phenomenon as the items are commoditized with lower and lower marginal cost).
We are addressing now the relatively untapped low price consumer market.
As the smartphone market will become saturated itself, the competition/innovation will concentrate on the next downsizing iteration: wearable computing, then ubiquitous computing and actually internally wearable computing when the younger generation, already accustomed to piercings and various neotribal body decorations, will adopt internal computing with micro surgical operations to “install” the devices.
On the other hand, looking back at what is behind the downsizing tide, the spectacle is not very enticing. Supercomputing had become very unimaginative, just piles of racks of xxxx lame PC servers. There is no minis segment anymore but a lesser pile of racks of xxx lame PC servers and the power user segment (workstations) is now fully neglected (oh, well yes, you name it with this rack of xx lame PC servers).
I don’t care of the power gamer PC niche and their shooting games with ever more realistic blood and body fluid splashs. Instead, I would like to see the heirs of Steve Jobs presenting along the iPhone5 an ambitious power user machine, mutatis mutandis, a 2012 $10’000 NeXT (Gosh, this very decorative NeXT cube I’m looking right now is already 22 years old). Of course forget it, this is so passé!
In 2020, my bleak prediction is that the PC segment will be fossilized too as the average people and software developers are not creative enough to really tap the teraflops (and soon enough the petaflops) at their disposal.
But chicks will be very fond of their wearable buds speaking the hour on a light touch, talking birds deployed at home as light alarms and a paraphernalia of gadgetry mixing jewelry with kitsch and recreative computing.
And yes, this is too long for a comment, sorry.
by vspyder
I’ve heard about the demise of the PC space for over TEN years. It still hasn’t happened and won’t happen in 2020…… This horse has been beaten to death and still it is alive……
As for your desire for a $10000 2012 version of NeXT there is NO market for this….. wishful thinking on your part I’m afraid…..
by Arcadium
I don’t know about you, but there are still plenty of enthusiasts who build very computation intensive computers and there are still plenty of uses for them (rendering, physics and engineering simulations, real-time analysis/etc).
An interesting side-effect of this commoditization of hardware is that distributed computing is real and outperforms equivalent best supercomputers.
For example the Bitcoin network currently crunches an estimated 180 Peta Flops while the top super computer in the world only does 16 Peta Flops!
On the flip side, for majority of users, we’re reaching a point where people don’t exactly see the benefits of extra computing power (excel isn’t running faster), for the next push it is likely we’ll need some form of Virtual Reality to cause a greater demand.
by Brian Mosley
Deary me, some sour nerds above ;)
The Raspberry Pi will inspire a generation. The whole point is, there will be kids now able to build their own supercomputer and looking to see what they can do with it… each one of those nodes has an easily accessible GPIO port and doesn’t have to be placed in a rack – the sensors could be dispersed throughout a lab or building.
Time to get creative!
Congratulations to the foundation and all the folks who are catching the excitement! I can’t wait to help my kids school get going with their own projects.
by Bri
I think the 13 amp power consumption is the real tell as to how much of a breakthrough this is. Most supercomputers are power hungry. Even if you could afford the hardware and software plus the support nessisary to run it, you’d need an electrical engineer and dedicated 50 and100 amp lines to run a relatively small one. This opens up supercomputing to very limited educational facilities.
by A4i
A single AMD FX-Series CPU will trash that supercomputer in every supercomputer test, and AMD system price will be way below £2,500. The problem with Raspberry Pi is that it’s single-core ARM11 CPU is notoriously slow. That ancient tech is not even supported in modern Linux OS core.
Nice start for Raspberry , but they must adopt some of the more recent ARM designs, not last century one.
by Marcos Marin
Very nice and dandy for an intellectual exercise BUT, how does it compare to an equivalently priced (or even cheaper) nvidia Tesla setup? SD cards are extremely slow… and you wont even need switches… (I guess you could use legos though, lol)
by AlexLoures
What are the requirements to be considered a supercomputer?
by Oto
I thought each unit has 256mb of memory which would bring the total memory to 16gb, the rest would be all virtual memory/storage no?
by Bri
A young Ray Kurzweil would have a blast with this! Supercool!!! Still a little pricey for average computer geeks.
by GAUSS
You could build a 16- or 32-node system and still get some cool results.