In the supercomputer universe, bragging rights go to the machine packed with the most number-crunching speed. And a spirited competition has raged for several years now between the U.S. and Japan for leadership in high-performance computing.

For the last two years, IBM’s BlueGene/L at the Lawrence Livermore National Laboratory kept the U.S. in the lead over a meteorological modeling machine developed by NEC called the Earth Simulator.

For those of you keeping score out there, Japan is about to take back the world speed record for computing it held earlier in the decade. The MDGrape-3 at Riken (formerly known as the Institute of Physical & Chemical Research) in Yokohama was clocked at a mind-boggling one quadrillion calculations per second. In industry-speak, that’s one "petaflop" of floating-point calculations per second.

After nearly four years in development and $9 million spent, the Riken machine is the first ever to accomplish the feat. It’s nearly three times swifter than BlueGene/L, the official No. 1 in an industry ranking called the Top 500 Supercomputer Sites. The MDGrape-3 wasn’t ready in time to qualify for the list which was released on June 27. It could top the next one, but the machine may be ineligible because of its specialized hardware. Here we take a look at the Riken machine and the global supercomputer race.

Should the U.S. government or researchers be worried that a Japanese supercomputer will soon be crowned the world’s fastest computer?

Not really. It is true that conceding the No. 1 spot in the supercomputer ranking would be a blow to the egos of U.S. scientists. But even if the U.S. loses the title, it still dominates the field, with 298 machines on the Top 500 list — more than any other nation. (Six of the top 10 supercomputers are U.S. machines.) Experts believe that the nation with the most machines near the top of the ranking generally has the most competitive economy. Trailing far behind the U.S. is Britain with 35 supercomputers, followed by Japan (29), China (28), and Germany (18).

What are the benefits of investing so heavily in supercomputing?

Supercomputers have a range of uses, from weapons development and scientific research to auto-safety testing and product design. Computer models can do the intense number-crunching needed for problems that can’t be cracked with experiments or that are too time-consuming for humans.

Meteorologists use supercomputers to predict climate patterns decades into the future by analyzing huge databases of statistics. Astrophysicists rely on the machines to test theories about the fabric of the universe, while engineers use them for crash-simulation tests or aerodynamics modeling of new cars.

What is the Riken supercomputer used for?

The MDGrape-3 will let scientists screen proteins that can potentially be used to make new drugs. The key is the machine’s speed: It will only take seconds to assess each protein. That’s a plus when you’re a pharmaceutical company testing tens of thousands of new chemical compounds. Researchers already use computers to determine whether such chemical compounds will bind to proteins in the body.

It takes a high-octane machine like the MDGrape-3 to show how tightly the bonds form, and how the molecules will stack up in 3D. For instance, the MDGrape-3 can show a researcher how a drug compound attaches to the HIV virus which causes AIDS. A subsidiary of pharmaceutical giant Merck (MRK) has already asked Riken for permission to try out the machine.

Building a supercomputer with just $9 million seems awfully cheap. How does that compare to the other high-speed machines?

No other supercomputer at the top of the rankings can muster so much calculating brawn on such a tiny budget. That’s partly because MDGrape-3 relies on fewer chips and less circuitry than rivals. It’s also because the chief scientist, Dr. Makoto Taiji, working with only two other researchers, had plenty of help from Hitachi, Intel, and NEC subsidiary SGI Japan.

Those companies supplied the hardware — Hitachi made the central processing unit, or CPU — and absorbed part of the cost of building the machine. One measure of the MDGrape-3’s ultra-efficient computing muscle is its cost per gigaflop (1 billion floating-point calculations per second), which Riken puts at $15. By comparison, BlueGene/L’s is $140 per gigaflop and the Earth Simulator’s $8,000.

As a measure of how far these machines have come, consider this: In the early 1990s, supercomputers cost about $1 million per gigaflop. To hold down costs, many computer scientists now use off-the-shelf chips and other parts. Even though MDGrape-3’s CPU was custom-designed, its other chips come from Intel (Xeon dual-core processors) and SGI Japan provided lots of the hardware and wiring.

The MDGrape-3 is also energy-efficient. While other supercomputers are notorious space-hogs (Earth Simulator requires a hangar-sized room) that need huge amounts of electricity, Riken’s machine occupies the space of a large walk-in closet and is an energy-sipper.

How long will it be before the Riken supercomputer’s record is broken?

Taiji, the Riken researcher, says he expects his record to fall within a few years. These days, the field is advancing at light-speed. For instance, Japan’s Earth Simulator, which hit the top of the charts in 2002, is now 10th and likely to be displaced from the top 10 by the end of this year.

Already several U.S. companies, including Cray (CRAY), IBM, and Sun Microsystems (SUNW) are preparing to create computers with petascale computing power.

Who keeps records of the world’s fastest computers?

Since 1993, the industry’s top computer scientists have tracked the world’s fastest computers in a twice a year (June and November) ranking known as the Top 500Supercomputer Sites.

The June ranking is timed to coincide with the International Supercomputing Conference, an annual industry get together in Germany. To make the list, machines must run speed-gauging software called Linpack. But speed isn’t the only goal, and the ranking’s editors frown on researchers who build machines that are fast but have no practical use or are later dismantled.

How do experts rate the MDGrape-3?

Alan Gara, chief architect for BlueGene/L at IBM’s T.J. Watson Research Center in Yorktown Heights, N.Y., had this to say: "It’s an unusual architecture. In BlueGene/L all chips can communicate with each other. In our largest BlueGene we have 65,000 nodes, with 130,000 processors. They didn’t need to do that. (MDGrape-3 has 4,808 chips.)

"They also built a processor that did only the type of calculations they need to do in astrophysics. So they built a specialized processor and a specialized network. It’s a good example. It shows how cost- and power-efficient you can be if you build for a specific applications. We can learn from it. They’ve set a benchmark of power performance."

While Horst Simon, associate laboratory director for computing sciences at Berkeley Lab and editor of the Top500 Supercomputer Sites, weighed in with this: "When we say 1 petaflop, it’s just a number. It’s the same as if you were to run 100 meters in less than 10 seconds. But it does mean something because it’s a barrier to break through. The fact is we’ve reached the petaflop threshold. Others will follow. In computing, a matter of three to four years can change things."

What’s the fastest possible speed of supercomputers?

That’s debatable. For now, it’s a safe bet that supercomputers will continue to get faster. Scientists are grappling with how to develop software applications and hardware architecture to squeeze more speed out of computers using less energy.

One future hurdle is the silicon chip that is the brain of these machines. Already engineers are reaching the physical limits of how many circuits they can cram onto a slice of silicon, though chip makers are experimenting with new technologies and materials for the next generation of microprocessors.

Did Riken receive government funding?

The $9-million budget of the MDGrape-3 came from Japanese government coffers. Riken developed the computer under a national project launched in 2002. Other countries have similar policies.

The U.S. Energy Dept.’s Office of Science spends about $230 million a year on supercomputer projects (though that’s only a fraction of total spending in the U.S. by industry titans such as Cray, Intel, AMD, Dell, Hewlett-Packard, and others). France, Germany, and China also pour millions annually into supercomputer research.