Subscriber Services
Subscriber Services
Complete Forecast
Search  Recent News  Archives  Web   for    
  •  Business Extra
  •  Financial Markets
  •  Technology
  •  Personal Technology
  •  Personal Finance
  •  Stock Options
  •  People & Events
  •  Drive
  •  Real Estate

Back to Home >  Business >


  email this    print this   
Posted on Wed, Sep. 29, 2004

IBM supercomputer may be world's fastest

Mercury News

IBM is expected to announce today that it has built a supercomputer that could qualify as the world's fastest machine.

The computer giant has built a version of its Blue Gene/L supercomputer that can beat NEC's 3-year-old Earth Simulator machine, according to Eric Krondstadt, the researcher in charge of the Blue Gene/L project in Yorktown Heights, N.Y.

IBM's newest version of the Blue Gene/L can sustain processing at 36.01 teraflops while the Earth Simulator has reached 35.86 teraflops. A teraflop is a trillion mathematical operations per second.

The new Blue Gene/L machine contains 16,000 PowerPC processors in 8,000 chips. Each chip has two PowerPC 440 processors, said Krondstadt, director of exploratory server systems at IBM.

He said IBM will deliver the final version of its Blue Gene/L machine, which will have 130,000 processors, to Lawrence Livermore National Laboratory in early 2005.

By comparison, the NEC supercomputer has 640 high-powered microprocessors. But the IBM machine is focused on delivering high-speed performance without consuming very much power. Krondstadt said the IBM machine will use 1/128th the power of the NEC machine.

``We're using a different design philosophy, which emphasizes low-power chips,'' he said.

It remains to be seen if the results will earn IBM the coveted title of producing the world's fastest supercomputer.

NASA is working with Silicon Graphics to build another contender for the throne by using 10,240 of Intel's Itanium 2 chips.

The Top 500 supercomputer list won't be out until November. The list is produced by researchers at the University of Mannheim, the University of Tennessee and the National Energy Research Scientific Computing Center at Lawrence Berkeley National Laboratory.

  email this    print this