IBM set to test the fastest computer in the world

The race is on for a crucial supercomputing breakthrough

The latest version of IBM's Roadrunner supercomputer is a hybrid machine that its builders expect will bust through the lofty petaflop barrier when it's tested this month. Akin to the significance of breaking the four-minute mile in the supercomputing world, the petaflop barrier is a goal that many computer makers, including Cray, Hewlett-Packard, Sun Microsystems and SGI, are shooting for.

Don Grice, chief engineer on IBM's Roadrunner project, thinks it's a race that IBM will win.

"We will break the petascale barrier," he told Computerworld at IBM's Poughkeepsie, New York facilities. "The only unknown for me will be what day it is. I don't think there's any technical reason we won't make it. The only hurdle left is persistence."

The new supercomputer will be used at the US Department of Energy's Los Alamos National Laboratory to work on national security problems, run annual tests of various nuclear weapons systems and predict long-term climate changes, according to John Morrison, leader of the high-performance computing division at Los Alamos. He says the system will also be used to study the universe and human genes.

"It will enable us to tackle problems we couldn't tackle before," he says. "Essentially, we'll be able to run a different level of problems. We'll be able to do calculations that we wouldn't even consider before." Morrison notes that the lab's contract calls for the new Roadrunner to reach the petaflop performance level.

Grice says the new machine would need a single week to run a calculation that the fastest supercomputer 10 years ago would have needed 20 years to complete.

If Roadrunner does break the petaflop barrier this month, it will mark the first time that IBM's BlueGene system hasn't held the highest position in the Top 500 supercomputer list since November of 2004, according to Jack Dongarra, a co-creator of the Top 500 list.

A petaflop is 1,000 trillion calculations per second. BlueGene runs at 478 teraflops, which is a trillion operations per second.

"It's exciting because it most likely will be the first computer to break the petaflop barrier," says Dongarra. "It's the next golden ring of computing. It's the next big marker. Today, all of the Top 500 supercomputers are at the teraflop rate."

Dongarra, a distinguished professor at the University of Tennessee, says that based on historical markers, the time is ripe to break a major computing barrier. Twenty-two years ago, a computer hit the gigaflop milestone and then 11 years ago the ASCI Red supercomputer was the first to hit a teraflop.

Dan Olds, an analyst with the Gabriel Consulting Group, says breaking the petaflop barrier won't just be a big deal for the industry but also for the researchers in various industries.

"Moving from teraflop to petaflop computing is a big deal," says Olds.

"Here's a comparison I like to use. The human brain is estimated to have about a 100 teraflop capacity. We only use maybe 10% of that — some of us even less. That means the Livermore computer is about 4.7 times faster than the brain, and the new Roadrunner system will be more than double that. It's seriously fast."

The new IBM supercomputer is the second incarnation of Roadrunner. The original Roadrunner, a cluster machine that can hit 70 teraflops, is in use at the Los Alamos lab. The older Linux-based system runs AMD Opteron chips.

This new version of Roadrunner, also runs Linux and gets its hefty power boost by adding the Cell chips, originally designed jointly by IBM, Toshiba and Sony for the latter's PlayStation 3, to the Opteron base. The hybrid supercomputer will use the Cell chips for massive calculations.

"We had done enough studies to see that it's one of the best computational chips in the world," says Grice, who first thought of using the Cell chip in a supercomputer. "It was built to do high performance computations for video games. The aspects that make it really good for gaming, made it really good for supercomputing. It's not running anything to do with the [operating system]. It's focused solely on calculations."

Steve Conway, an analyst at IDC, expects supercomputer manufacturers to design a lot more hybrid machines down the road.

"It's really an interesting design, combining x86 processors with a processor from the domain of video gaming," he says. "It fits into a much broader trend toward hybrids that goes beyond supercomputing. We believe this trend will accelerate."

The new Roadrunner uses 3.9MW of power, which Grice notes is enough to power 39,000 100-watt light bulbs. It has 6,948 dual-core Opterons on IBM LS21 Blades, as well as 12,960 Cell processors on IBM QS22 blades. The machine, which has 80TB of memory, has 296 IBM BladeCenter H racks. It takes up 6,000 square feet, uses 57 miles of fibre optic cable and weighs in at 500,000 pounds.

Once the IBM technicians finish testing it, they'll pack it up on 21 tractor trailer trucks and move it to Los Alamos in New Mexico, where they'll reassemble it and test it all over again.

IDC's Conway says the next several weeks should prove interesting. The Top 500 list of supercomputers will be updated on June 17 at the International Supercomputing Conference in Dresden, Germany. He says any company close to hitting a petaflop will be testing their supercomputers to try and make the list. "It has symbolic importance," says Conway. "It's like the four-minute mile. It gives people confidence that they can hit 10 petaflops and 1,000 petaflops."

Last week, NASA and partners SGI and Intel announced that they are working on a supercomputer that they expect will pass the petaflop barrier next year and hit 10 petaflops by 2012.