GPU : A Giga@lobal Processing Unit ?

Final thoughts




There is a spacious plain filled with alacreous interconnected processors, a strange syntethic world somewhere in cyberspace, between the transport and application layer.

Processors appear small and slow compared to their big brothers sitting in the computing centers and the network they form is a web. A closer look at the web reveals its dynamic structure: it seems as if it would be spun by invisible spiders again and again and again.

Processors talk to each other using a common protocol that allows them to both establish new and softly release unused connections. Over these connections they exchange jobs, files, search and computation results. The processors have a lot of jobs waiting in queue to be accomplished. They search for music files, optimized golomb rulers (used in radioastronomy) and huge prime numbers. They try to factorize RSA numbers and perform repetitive tasks on huge chunks of data, e.g. they Fourier transform signals recorded from a radio telescope in search for life outside the solar system. They perform automated docking of flexible ligands to a protein’s binding site in search for a cancer cure.
Sometimes they even manage to break a cipher!

Processors are only aware of their network’s neighbors. What perhaps sounds incredible is that there is no need to spend hours drawing maps of how they are interconnected. Why draw a map if some moments later nothing will be the same? The other reason is that they have to deal with more important problems.

Some of them suddenly disconnect because of power failures, some others die hard trying to find infinity in a divide-by-zero operation. Others get crazy because they lose control of their memory. Malicious beings, nodes created with the purpose of destroying the anarchic network try flooding connections with useless packets. Or they could eat all packets, becoming a black hole. In addition, these guys don’t respect the protocol; they simply speak in horrendous slang.
It's a hard world, in which chaos breaks down with determinism.

But sometimes, like a meteor in the dark sky, another modest processor comes up with some useful connections, after finding out where the moving underground network went. It was helped by the benign host catcher, which keeps a list of living nodes !

The new processor is small, slow but pigheaded, and it’s willing to help with its small draws (despite seeing the whole picture, of course). Through its connections, streams of bits flow like new hope for the dying Gnutella network.



Traybar of a computer fanatic owning a Pentium III one gigahertz. It is still possible to work normally.



CPU load after starting a GPU Job


The logo was designed by Mark Grady. Graphics are by David A. Lucas.
© 2002-2003 the GPU Development team

Page generated in 11.68 milliseconds by PHP 4.4.3-dev on seeschloss.free.fr.