Intel Corp has been staking out its ground in supercomputing into the late 1990s and describing its somewhat hybrid vision of machines capable of producing one (American) trillion arithmetic operations per second. Intel sees the TeraFLOPS goal being achieved by using a combination of high speed microprocessors and massively parallel processors. It reckons that by the mid-1990s, TFLOPS computing will be a reality and that the impact on scientific advancement will significant. In areas such as space research and global climate modelling, where it is impossible physically to test theories, scientists rely on computation as a substitute to experimentation. And the faster they can compute, the quicker they can test their theories. A global climate modelling project that would take a year to compute on current GFLOPS computers, would take a TFLOPS machine just one day. This type of modelling is just not attempted at the moment as it takes far too long. As part of the quest for TFLOPS machines, Intel has embarked on a $27m joint research and development project with the US Advanced Research Projects Agency. The Touchstone project, (CI No 1,338), revealed the first fruits of its labour, the 7.6 GFLOPS Gamma machine, back in December 1989 (CI No 1,517). Last month the stepup Delta machine began internal testing. It runs at 33 GFLOPS and comes with up to 1Gb memory. Intel will formally announce the machine next month. The final machine in the series perhaps before the TFLOPS goal is reached, is Sigma, scheduled for the end of 1991. Sigma should run at 150 GFLOPS and support up to 100Gb. All three machines are based on the 80860 chip.
