Goldin: Computing must move beyond silicon
Connecting state and local government leaders
JUNE 8—The nation needs to look beyond silicon to unconventional computing technologies, NASA administrator Daniel S. Goldin said this week.
By Patricia Daukantas
GCN Staff
JUNE 8The nation needs to look beyond silicon to unconventional computing technologies, NASA administrator Daniel S. Goldin said this week.
The space agency chief spoke at a Silver Spring, Md., forum highlighting supercomputer maker SGI's collaboration with weather and climate research in federal laboratories.
Goldin said that when he assumed the top NASA post nine years ago, researchers were pushing computer makers to achieve 1 trillion floating-point operations per second. But he said he is convinced that NASA will need several tens of thousands of TFLOPS, or tens of petaFLOPS, to meet future needs.
Design of complex systemsaircraft, satellites and the International Space Stationis still done empirically, but people are surprised when a design fails, Goldin said.
'You do the best certification process you can, and then it's trial and error,' he said.
Systems that can execute 40,000 TFLOPS, or 40 petaFLOPS, will be necessary to do the enormously detailed structural analysis and flight dynamics engineering that Goldin envisions. Silicon chip fabrication cannot maintain the necessary rate of increase in computer performance, he said.
He called for spending $100 million on such unconventional approaches as nanotechnology, which would build ultra-small devices one atom or molecule at a time.
Goldin urged government researchers to advance their data analysis techniques before their observations and computer simulations turn into 'data morgues.' Last year NASA collected 330T of data, more than in all of the past 40 years, he said. In the future, the agency will have petabytes or exabytes'millions of terabytes.
He called weather and climate prediction 'probably the single toughest problem we have.'
Many research groups have developed climate prediction models but can't share data sets easily, said James Fischer, manager of the Earth and Space Science Project at NASA's Goddard Space Flight Center in Greenbelt, Md.
The space agency has selected nine proposals for building advanced climate-change models that will interoperate, Fischer said. The project will spend $18 million to get such models into production by 2004.
NASA Ames Research Center at Moffett Field, Calif., has achieved large performance gains by porting codes designed for vector supercomputers from Cray Inc. of Seattle to SGI Origin 2000 and Origin 3000 platforms, Jim Taft of Ames' Terascale Applications Group said.
The center is installing an Origin 3000 that will be the first of its kind with 1,024 processors working as a unit, Taft said.