Top computer hangs on to its title
IBM's BlueGene/L computer at Lawrence Livermore National Laboratory (LLNL) in California, has once again been crowned world champion by the TOP500 list of the fastest supercomputers used for scientific applications. Ned Stafford digs into the story behind
Just how fast is BlueGene/L?
This giant among giants has 131,072 processors and a computing speed of 280.6 terraflops per second (1 teraflop equals 1 trillion calculations or 'floating point operations').
Far behind in second place is an IBM computer dubbed BGW (Blue Gene Watson), with the same technology as BlueGene/L but fewer processors and a computing speed of 91.29 TFlops. Tenth on the list is the NEC-built Earth-Simulator in Yokohama, Japan, at 35.86 Tflops, which held the world title from the time it went into operation in 2002 until 2004, when it was dethroned by an earlier version of the current champ. The computer at the bottom of the list, the five-hundredth fastest, performs at a speed of 2.026 TFlops.
Wait just a nanosecond. I'm having trouble processing all this teraflop stuff. What does that mean?
If everyone in the world, 6.6 billion people, each had a calculator and performed a simple calculation every 5 seconds, it would take the entire planet nearly 60 hours to do the same number of calculations that BlueGene/L can do in 1 second.
Why does anyone need anything that fast?
BlueGene is used for nuclear weapons research. But other supercomputers do everything from stock-market analysis to climate forecasting.
What is the TOP500 list?
The list is compiled twice a year (June and November) by a team of representatives from the University of Mannheim in Germany, the University of Tennessee in Knoxville and Lawrence Berkeley National Laboratory in California. The most recent list, issued this week at the International Supercomputer Conference (ISC) in Dresden, Germany, can be found on the TOP500 Web site.
What is the technology behind these computers?
There are some familiar big names in the supercomputer market. IBM is the dominant supercomputer vendor, with Hewlett-Packard coming second. Intel microprocessors are used in 301 of the 500 systems. But Advanced Micro Devices (AMD) is growing quickly, Simon says.
In software, Microsoft is at the bottom of the list, with only two supercomputers using its operating system. The software king is Linux, used on more than 70% of the total.
I want one. Can I have one?
Only if you're rich. The most powerful supercomputers could cost US$100 million or more, while low-end models come in at $1-2 million. It is also possible to link a lot of PCs together in a cluster that has supercomputer capacity; about 1,000 powerful PCs could make the bottom of the TOP500 list. "There are recipe books out there telling you how to do this," Simon says.
You might be interested to know that today's average notebook computer is faster than the supercomputer that made the bottom of the TOP500 list in 1992. So you may already own the equivalent of a vintage supercomputer.
Will quantum computers be able to surpass today's supercomputers?
Simon says quantum computers hold great potential, but that he believes they would be used for performing "completely different types of calculations" than supercomputers. Comparing the two types would be like comparing apples and oranges.
How much longer will BlueGene/L hold the top spot?
While computer chip capacity is doubling every 18 months or so, often referred to as Moore's Law, supercomputer capacity is doubling every 11 to 12 months.
Simon thinks BlueGene/L's current lead is so great that it will remain champ at least until the November 2007 TOP500 list, maybe even until November 2008. But the next TOP500 champ could shatter the current record. "I would expect a big jump to 500 or 600 or 700 terraflops, maybe even to a petaflop," he says; that's one thousand teraflops. "Petaflop is the next magic number."
Visit our newsblog to read and post comments about this story.
User Tools [+] Expand
If you need help or have a question please use the links below to help resolve your problem.