INTEL, best known for making the chips in personal computers, has unveiled a super computer capable of performing one trillion operations veer second.
The new machine, created for the US Energy Department, performs 1.06 teraflops, or 1.06 trillion floating point operations per second. This shatters the previous computing record held by a Japanese university computer designed by Hitachi by more than 250 per cent, according to Intel.
"Today's accomplishment is computing's equivalent to breaking the sound barrier," Intel's chief executive, Craig Barrett, said at the announcement in Santa Clara, California. "We could be at the threshold of robust scientific discovery, triggered by access to teraflop level computing performance."
The $55 million super computer will be used by US government scientists to simulate nuclear weapons tests now banned by an international treaty, as well as in weather forecasting and in genetic and space research, officials said. The machine is able to process in 15 seconds what the average desktop computer takes two days to crunch.
Using "massively parallel computing", researchers linked up 7,264 Pentium Pro processors to work in concert, creating a new supercomputer. The project shows for the first time how common microchips can be built into a massive machine, instead of having to specially design elaborate and costly chips for traditional supercomputers.
Housed in 577 refrigerator sized cabinets at an Intel site in Beverton, Oregon, it was first used on December 11th when it performed 6.4 quadrillion (a quadrillion has 15 zeros) calculations in 80 minutes, according to Intel vice president Edward Masi.
The final version of the new supercomputer will have 9,000 200 MHz Pentium Pro processors and will be able to operate at 1.4 teraflops, Masi said.
Given its high price, no orders have yet been placed for the supercomputer, but the US Energy Department said it would allow businesses and scientists to purchase spare time on the computer.
The new computer uses the Linpack measurement method, the most widely recognised single benchmark for measuring sustained floating point performance of high end computers.