"The models are sufficiently accurate to give powerful qualitative insights to our colleagues to aid their design process", Mr. Smith stated.

The chemical interactions the computer is dealing with are exotic to say the least, and grasping their complexity is probably best left to the engineers who create them.

Mr. Smith said the models sought to map interactions ranging from a few hundred thousand atoms to about 1 million, mostly in scenarios in which measuring outcomes in a direct experiment was almost impossible. The models could, for instance, tell a biomolecular nano-engineer what is likely to happen if a particle is used to deliver DNA into a cell to slow development of proteins that cause Parkinson's disease.

They are also being used in ambitious projects to come up with environmentally friendly alternatives to fossil fuels. For instance, they are developing predictions on what sort of material might help titanium-dioxide use solar energy to split water into hydrogen and oxygen more efficiently.

The university had a budget of about $1million to establish the system, with more than $500,000 allocated for hardware, Mr. Smith said.

The use of common commodity-based Intel hardware for supercomputing is a relatively new concept. Historically, the territory has been dominated by proprietary system makers such as Sun Microsystems and IBM.

Specialist system architecture would not have enabled the institute to tailor its system to suit the applications, Mr. Smith said.

The Dell cluster was stratified so that some processors in the system had more memory and disk space allocated. It was preferable to buy 300 servers and mix and match than buy an out-of-the-box high-end proprietary system.

"If I wanted to bundle up all the memory I have on all of those 300 boxes and I wanted to make it available for one calculation, I'd have to buy a large memory machine, and that kind of architecture is very expensive", Mr. Smith stated. "I wanted the same amount of memory for one calculation I could only afford to buy maybe 10 or 29 processors, because of the way the cost of architecture scales, whereas if I buy 300 boxes with a smaller amount of memory and a hard disk for each one the costs are a lot less", he stated.

That is not to say Sun and IBM could not supply the kind of hardware needed for the job. They also offer servers in blade arrangements with standard Intel and AMD processors.

In this case, the university was capitalising on a good offer from Dell when it took the project to tender.