Future Champs

Blue Gene

IBM is involved in the building of a new supercomputer, that will be the new speed champion. This project, known under the name Blue Gene is actually about building three different supercomputers – Blue Gene/ L at 180 TeraFlops, Blue Gene /P at 1000 (PetaFlop)TeraFlops and Blue Gene /C, also at 1000 TeraFlops. The project was started in 1999 with a budget of USD 100 million.

Blue Gene/L is to be used for studying protein folding. Proteins are made up by stringing together amino acids into long chains. These chains fold about each other in many ways, and the way they fold determines how they function. Improperly folded proteins can poison the cells around it. So, understanding the way proteins fold may hold the key to curing many diseases like cancer. But this requires enormous amounts of computing power.

When completed, Blue Gene /L will have 65536 nodes interconnected by three different networks (most supercomputers have multiple networks linking their nodes, each performing a different function). The first is a three dimensional torus, connecting all nodes and forming the communications backbone. The second is a Global tree network, which helps aggregate arithmetic operations across nodes. And the third is Ethernet, built into every node ASIC, that is to be used for booting, I/O, diagnostics, etc.

Bluegene Architecture: Bluegene is to be built on a system on a chip concept, with memory being on the chip along with the processors. Disk storage is separate

Bluegene /L is to be built from embedded PowerPC chips and is expected to operate at 200
TeraFlops.

Linux Power

Another supercomputer is to be built by HP at the Pacific Northwest Laboratory in Washington for the US Department of Energy. This will be powered by 1400 Itanuim processors and will run Linux. Set to achieve 8.3 TeraFlops of computing power, it will cost USD 24.5 million.

Ultra small?

In the supercomputing big is normally better. But at the other end of the spectrum, we have the world’s smallest cluster computer. Built by the US Department of Energy’s Sandia National Laboratories, this cluster is actually portable! 

What you see in the picture is a 4 CPU cluster computer running Linux. The cluster uses PC 104 standard based off the shelf embedded components stacked on top of one another. A detailed how-to with pictures on how to build this machine is available at
http://eri.ca.sandia.gov/. 

Can this machine deliver supercomputing power? No. This one is to be used for road shows, demonstrations, proof of concepts and even work on the move rather than for real life number
crunching.

Krishna Kumar