How To Guides Org

Inspiration: Building a Cheap Supercomputer in 2014? Is a “Cheap Cray” Alternative Possible?

I’ve been pondering over two questions. 1) Is it possible to build a cheap supercomputer like a cheap Cray alternative and 2) If yes, what would you do with it other than calculating PI and setting a new record?


What Is A Gigaflop?

Alright, you probably know what Ghz are (Giga-hertz). The Giga stands for 1024 Megabytes and is what we work with on a regular basis in the consumer market: It is a measurement of (clock) speed, so that consumers know how fast a CPU is. Most modern PC’s include CPU’s with a clock speed between 1 and 4 Ghz. This AM3+ AMD CPU for example has 4 Ghz and would set you back roughly 200 bucks.

Gigaflops, unlike Gigaherz, don’t really measure the speed of a CPU (clock speed) instead flops measure the number of floating point operations. To fully understand what that means, you should know what floating numbers are.

What Are Floating Numbers? Is Pi One?

If you start programming a little, you will sooner or later encounter integers and floats. Floats / Floating point numbers are real numbers like Pi 3.14159, 1/2 and so on.

When you’re multiplying PI with PI (example) then you are performing a floating point operation: A flop. If you measure how many of those operations you can do per second you know how to compare the best CPU’s out there: In Gigaflop/s.

2007: Cheap Supercomputer That Could Do 26.25 Gigaflops

First, let’s talk a little about the past. In 2007, some guys over at clustermonkey build a $2500 cluster for $100/GLFOP

Fast-Forward: 2012 Supercomputer Raspberry Pi

You can find out more about it over at http://www.southampton.ac.uk/

What Is The Current PI Record?

If you managed to get the funds for a supercomputer and have the time to let it run for a while, it’s still very unlikely that you will manage to break the PI calculation record. In 2011, a team from Yahoo used a 1000 computer cluster to calculate the most PI digits.

In 2013, Ed Karrels managed to set a new record with the help of NVIDIA hardware CUDA, a parallel computing platform

(Image Source: karrels.org/pi/)

Why Consider Building A Supercomputer?

I encourage everyone who’s interested to consider building one and think big, really big. The sky is the limit here. If the human species wants to find a way to detect asteroids and avoid getting wiped out we should start building lots more supercomputers.

Supercomputers could help to analyze sky patterns and identify possible asteroids or they could help to solve some of the biggest problems we have right now: The water crisis, over-population and of course the energy crisis. But there are loads of big problems that require solutions and some of them require massive calculations.