Virginia Tech using 2.3 GHz Xserves
"These aren't your run-of-the-mill Macs; each sports a 2.3 GHz IBM PowerPC 970 processor, which isn't available to the Little People. "
They are hoping to reachthe #3 spot for supercomputing power.
Edit: Sorry for bad link here is the whole article:
Andrew Kantor: CyberSpeak - Inside -really inside - one of the world's fastest computers
Yesterday I stood inside what is probably the world's third-most-powerful supercomputer - the Terascale Computing Facility at Virginia Tech. (I say "probably" because the testing won't be done till next week - the Department of Defense (news - web sites) is using it right now and can't be interrupted.) It's composed of 1100 Macintosh (news - web sites) G5 computers running in parallel.
These aren't your run-of-the-mill Macs; each sports a 2.3 GHz IBM PowerPC 970 processor, which isn't available to the Little People. But before you Mac people start giving each other high-fives, you should know that the university didn't say, "Hey, let's make a supercomputer out of Macs." They were interested in that PowerPC processor, which IBM happened to sell to Apple for those Macs. (All right, one high-five is allowed.)
I was told this by Srinidhi Varadarajan, an assistant professor of computer science who took me on a tour of the Brobdingnagian place. (Everyone pronounces his name "Serenity," which I'm sure is not quite accurate, but he doesn't seem to mind. If I could say, "My computer's more powerful than yours" to all but two people in the world, I wouldn't care what people called me.)
'Scuse me while I whip this out
Everything about the computer is big. The Macs themselves fill about six 40- or 50-foot rows of racks, each about seven feet high plus a monster cooling unit on top. (It was comfortably cool in there, but Srinidhi said that it would only take a matter of seconds to get into the 100s if the A/C cut out.)
That A/C works with the same coolant used in your car, but the coolant is itself cooled by water - water pumped in at 750 gallons a minute. Think Alabama in the '60s.
The supercomputer will not be winning the Jimmy Carter Energy Conservation Award: Its breaker panel can handle up to 1.5 megawatts of electricity, although the system only uses about a third of that. For comparison, if you have 200-amp service in your house, that's 24,000 watts (200 amps x 120 volts). At 1.5 million watts, this has more than 60 times the capacity.
Monster computer, monster power, and monster backup: A room full of batteries gives the computer what Srinidhi called "the largest uninterruptible power supply in the State of Virginia." As a bonus, outside sits a 1.7 megawatt diesel generator that kicks in if the batteries run low.
All that power powers a lot of power: The supercomputer can sustain 10 teraflops of computing, and actually perform significantly higher than that at its peak. For many of you, I realize, that means nothing. So a brief digression.
A flops is a "floating point operation per second" (note that the singular is "flops," not "flop"), and is one measure of computing power. The more flops you computer can perform, the faster it is.
A nicely powerful home computer performs at a few gigaflops, that is, a few billion operations per second. So at more than 10 teraflops, Virginia Tech's computer is about a thousand times faster.
And what, pray tell, is the purpose of all that power? You might think it's a matter of boys being boys: Build the biggest, fastest, tallest, etc., just for the sake of it. And I think there's some of that involved. But there's a good use for all that power. Actually, several. But we need another digression to explain that.
I knew you were gonna say that
In theory, if you know enough about something now, you can figure out what it's going to do later. So said Isaac Newton back in 1687 when he published The Mathematical Principles of Natural Philosophy (aka The Principia). In the early 1900s quantum mechanics came along and said, "Well, not exactly," but in general Newton's laws hold up for the things we can see.
So if you know enough about, say, the weather at this moment, you can predict where it will be tomorrow.
The problem is, it's really, really hard to know enough; you bump into chaos theory. Chaos says, among other things, that tiny differences now turn into huge differences later. A difference in a fraction of a degree in temperature today can profoundly affect the weather in a week. (That's the too-often-quoted butterfly effect: Whether or not a butterfly flaps its wings in China can determine whether it rains in New York. A butterfly played a similar but more dramatic role in Ray Bradbury's classic story "A Sound of Thunder," which I see has been made into a movie that looks to be a travesty of the story.)
This is why it's well-nigh impossible to predict the weather more than a few days in advance: No matter how much we know now, there are too many little things we don't know that can change things dramatically - a degree here, a butterfly there.
Still, the more information we have can work with, the more accurate our predictions can be. That's where supercomputers come in.
Weather models are a big use for the kind of power Virginia Tech offers (and, by the way, offers at a much cheaper rate than other supercomputing centers). The atmosphere is a complex thing - four dimensions of molecules that interact in incredibly complex ways. Sure, we can make general predictions, but those aren't very useful. Whether a hurricane will turn north here or here is a big deal. Ask someone in Florida.
Then there's defense. Or, more accurately, offense. The DoD uses supercomputers to model nuclear explosions. Thanks to a variety of treaties, we can't actually test our nukes by setting them off (although the Bush Administration plans to pull out of some of those agreements), so we test them virtually. Like the weather, though, a nuclear explosion is a fairly complex thing, and modeling what happens when the button is pushed requires a lot of raw power.
Drug companies, too, need supercomputers for a few things. Molecular modeling is one. Many drugs do their thing by connecting with a particular molecule in your body. But it's a lock-and-key thing; the drug has to be just the right shape to fit its target. Computer models need to try millions upon millions of combinations to see what will fit. And some of those molecules fold in different ways, so that add millions of additional possibilities. Without a supercomputer, the process could take years longer.
These are the kinds of things that only a supercomputer can do. There are other problems that can be solved in other ways, such as with distributed computing. Projects such as [email protected] send them to thousands of home computers, which together might have as much power as a supercomputer. But not every problem can be easily split up that way.
As our tour was concluding, I asked Srinidhi something I always wondered: "What kind of interface does a supercomputer have? What operating system?"
"Oh, it's Mac OSX," he said. (In other words, it's a version of Unix (news - web sites).)
So the most powerful computer in academia has as it desktop the same screen someone buying a Mac in Circuit City has? "Exactly," he said. But, he pointed out, "we turn off the graphical interface so you have a nice command line."
Andrew Kantor is a technology writer, pundit, and know-it-all who covers technology for the Roanoke Times. He's also a former editor for PC Magazine and Internet World. Read more of his work at kantor.com. His column appears Fridays on USATODAY.com.