News Blog

NASA, Intel, and SGI announced today that they are collaborating on a groundbreaking initiative that promises to vastly improve performance of the space agency's supercomputer operations, "enabling them to push the limits of scientific discovery."

The space shuttle Atlantis

(Credit: NASA)

Under a joint project dubbed "Pleiades," the three partners plan to develop a modeling and simulation system of unprecedented speed and capacity in the nation's space program. Specifically, they hope to produce computational performance of 1 petaflop (a quadrillion operations per second) by 2009 and 10 petaflops by 2012.

What does that mean, exactly? A task that would take six months to complete on a PC could be done in 1 hour by the Pleiades system, according to an Intel spokesman. Or, put another way, it's analogous to a 6-hour cross-country flight taking just 1 second.

For NASA, the benefits of the initiative are far more than just theoretical: Its predecessor, Project Columbia, allowed engineers to simulate emergencies in time to avoid space shuttle disasters. The partners say there is a green element to Pleiades as well, a goal to design new aircraft that are 70 percent more fuel-efficient than today's models and make only a fraction of the noise.

Purdue University on Monday plans to install a new supercomputer on its campus in a single day, in an effort it's calling an "electronic barn-raising."

Approximately 200 employees will help build the machine, which will be the size of a semi trailer and is expected to have a peak performance of more than 60 teraflops, which means it would be able to perform more than 60 trillion operations in one second, according to a Purdue press release.

Once it's completed, Purdue's supercomputer still won't top the list of the 500 fastest supercomputers, though the university says the machine will rank in the top 40. IBM's BlueGene/L at Lawrence Livermore National Laboratory currently claims the crown for the world's most powerful supercomputer, but turnover on the list is rampant.

Organizers of the campus event posted a spoof trailer on YouTube called Installation Day that takes a jab at the film Independence Day.

Yahoo announced Monday that it has joined forces with the Pune, India-based Computational Research Laboratories (CRL) in order to support research in cloud computing, a way to outsource hardware and software to service providers rather than host it locally. Under the terms of the agreement, researchers will be able to use the EKA, a supercomputer owned by CRL that contains 28 terabytes of memory, 14,400 processors, 140 terabytes of disks, and a peak performance of 180 trillion calculations per second.

According to CRL, EKA is the world's fourth fastest supercomputer.

"We are excited to partner with Yahoo to advance cloud computing research in India as it opens up a new arena of exciting opportunities," Gautam Shroff, a member of CRL's steering committee, said in a statement. "We are initiating dialogue with leading Indian academic institutions to collaborate on research using cloud computing."

At the core of Yahoo's involvement is its role in the Apache Hadoop project, for which it opened an open-source research and development center last November. Yahoo and CRL's announcement was timed in conjunction with the inaugural Hadoop Summit, sponsored by Yahoo and the National Science Foundation-funded Computing Community Consortium.

MENLO PARK, Calif.--It got delayed a few months, but a new, somewhat unusual supercomputer from Sun Microsystems will get formally unveiled next week.

The Texas Advanced Computing Center (TACC) at the University of Texas will dedicate a Constellation System from Sun on February 22, said John Fowler, executive vice president of systems at Sun. He was speaking at the company's global media summit here Wednesday. (Technically, the computer started running earlier this month: the dedication is sort of like the official coming-out party.) After TACC, Sun hopes to start selling Constellations to more customers.

The linchpin of Constellation is the switch, the piece of hardware that conducts traffic among the servers, memory, and data storage. Code-named Magnum, the switch comes with 3,456 ports, a larger-than-normal number that frees up data pathways inside these powerful computers.

"We are looking at a factor-of-three improvement over the current best system at an equal number of nodes," said Andy Bechtolsheim, chief architect and senior vice president of the systems group, in June about the Constellation concept. "We have been somewhat absent in the supercomputer market in the last few years."

Sun had hoped to launch the TACC system in October, but it ran into a variety of technical problems. First, AMD delayed the Barcelona processors that go inside the computer. "We got a special run of chips from AMD to make our commitments," said Fowler. Sun will later release more standard Barcelona servers when the chips become available.

But it wasn't all AMD, Fowler said. Constellation also sports a new type of cable, invented by Sun, which comes with three connections per cable. Manufacturing these cables, and then snaking them around the TACC center to link up computers, proved tougher than expected, he said.

Technical glitches also popped up with the Magnum switch. (Even though Sun did have technical problems with its own computer, Fowler said that third-party suppliers would provide Sun financial compensation for delays, the normal arrangement in these types of contracts.)

The TACC system will provide a peak performance of around 500 teraflops, or 500 trillion operations a second, and can be increased. It will be made up of 82 Sun blade racks stuffed with servers, 2 petabytes of storage, said Fowler. The whole system will fit inside a mid-size conference room but provide more computing power than all of the supercomputers the National Science Foundation has today.

The architecture will also allow Sun, according to the company, to challenge IBM in the rankings for the world's top supercomputers. IBM has dominated the supercomputer rankings with a series of Blue Gene systems for the last several years.

Microsoft released the public beta of Windows HPC Server 2008 for running large computing clusters, part of its plan to creep into supercomputing.

HPC Server 2008 is the successor to Windows Compute Cluster Server 2003. HPC Server runs on the individual servers in a cluster and also comes with higher-level software that coordinates all of the members of a cluster. Microsoft says it achieved a 30 percent improvement in Linpack, a commonly used supercomputing benchmark, on its production cluster with 2,048 processor cores.

HPC is also being used on a 1,151 node cluster at the Holland Computing Center at the University of Nebraska.

But not so fast! Rival Red Hat and Platform Computing on Tuesday announced Red Hat HPC Solution, which combines Platform's Open Cluster Stack1 with Red Hat Enterprise Linux. Effectively, this allows computer architects to do the same things, but with different software.

Although Linux is little seen on desktops, Linux and Unix are far more dominant in this market.

The announcements, like yesterday's list of the world's 500 fastest computers, took place at SC07, a major supercomputing conference taking place in Reno, Nev.

Later in the week, expect to see an announcement that simulations conducted by a group of German scientists on a 4,420 node cluster still have not been able to determine how the Cal-Neva can deliver a 99 cent breakfast and still turn a profit.

Could either of these Rubik's Cube prodigies solve the cube in 26 moves? Leyan Lo talks strategy with Shotaro Makisumi at a competition in January 2006.

(Credit: Daniel Terdiman/CNET News.com)

Editors' note: This blog initially misspelled the name of a record holder for solving a Rubik's Cube. He is Leyan Lo.

Clearly, I've been doing something wrong.

Since the early 1980s, when I got my first Rubik's Cube, I've never been able to solve it. Oh, sure, I got one side done, and maybe even two. Or, I could break the thing open and put it back together in its original, solved position.

But now, according to the BBC, a supercomputer has determined that a Rubik's Cube is solvable in less than 26 moves, regardless of the starting position. So, boy, don't I feel dumb?

It turns out, thanks to research done by Northeastern University graduate students Daniel Kunkle and Gene Cooperman that that's all it takes to solve one of the famous toys. Duh.

Yet, I wonder: Could the computer that proved this beat the likes of Leyan Lo, who early last year set the world's record of 11.13 seconds? I sort of doubt it. After all, have you ever seen a supercomputer try to turn a Rubik's Cube?

Making a supercomputer used to require teams of dedicated scientists, millions in federal research grants, and lots of specialized components that took years to design.

Thanks to clustering and other advances, a group of well-trained grad students can build one from off-the-shelf parts. As a result, the rankings in the Top 500 Supercomputers list changes more rapidly than the standings on Dancing with the Stars.

So with commoditization, what's the most expensive thing? The cooling system? The processors?

Weirdly, it's the memory, says Andy Bechtolsheim, senior vice president of Sun Microsystems' systems group and a co-founder of Sun, who revealed the company's Constellation System this week.

"The dominant cost of the systems is memory, DRAM, believe it or not," he said. A typical supercomputer will require millions of DRAM chips. DRAM drops in price like crazy--the average selling price of DRAM dropped 35 percent from December to April. But millions of chips are millions of chips.

The cables aren't cheap either. In fact, they cost more than the silicon inside switching systems.

Electricity is a problem, too.

"The cost of electricity is now showing up in the bids," he said. It varies by geography. Sandia's Oak Ridge lab in Tennessee has an advantage in this area because it's located in a hydroelectric hotbed.

The BBC on Friday ran a piece about a team from IBM and the University of Nevada that has created a model simulation of half a mouse brain on a BlueGene L supercomputer. Well actually, it simulates only a portion of half a mouse brain's cortical cells. And actually, it simulates only a portion of their functionality. And it was able to run at only one-tenth the processing speed of the real thing.

It was still a major accomplishment, though, since they were able to overcome problems surrounding the "tremendous constraints on computation, communication and memory capacity of any computing platform," the BBC reports.

The blogosphere was overrun with clever headlines, and Jamais Cascio even posted a shot of Pinky and the Brain--the lab mice who plan to take over the world on Warner Brothers' Tiny Toons. We'll let him deal with the copyright people on that one.

Blog community response:

"I, for one, welcome our new cyber-mouse overlords!...(T)his is a simulation of a cortical network with the size, link complexity and signal activity of a mouse brain, but without the structure--so, arguably, it isn't a really a simulated mouse brain, but a functional platform upon which a mouse brain sim could run. Depending upon your perspective, this is a minor quibble or makes all the difference. "
--Open the Future

"Mice still too slick for supercomputers...Humans may have the ability to destroy the world, fly to the moon and build an information superhighway, but for some reason we still can't figure out how a mouse thinks...There's just one hitch: The supercomputer is lagging behind half a mouse brain on a one to ten ratio. The mouse can literally run circles on the CPU with ease."

" I still believe that no matter how good a brain simulation a computer can create, it will never become an independent mind. Animal brains are electrochemical systems and simulating just synapse firing patterns will never approximate the effect of neurotransmitters and hormones on the brain."
--Dvorak Uncensored

"That's a lot of computer firepower for one second of mouse brain activity. So, it's going to take a lot of effort to attain, say, a whole week's worth of mouse brain...By the way, my guess is that one second of mouse thought involves food, sex or survival. "
--Trouble Ain't Over

The U.S. Army will spend $105 million over five years on a new supercomputing research center based at NASA Ames Research Center in Mountain View, Calif. Stanford University will lead development of the research center, with the help of engineers and scientists from NASA, Morgan State University, New Mexico State University and the University of Texas.

The goal of the new center is to test, via computer simulation, various new military applications and hardware, such as wireless communications on the battlefield, advance warning systems for biological warfare, and lighter materials for army vehicles. The research could also prove beneficial to supercomputing research itself.

"Using the most advanced high-performance computing resources, a research center of this magnitude has great potential for innovating technology," Charbel Farhat, Stanford mechanical engineering professor, said in a statement.

The grant from the U.S. Army, which began its supercomputing research center program in 1989, includes an optional renewal for another five years and $105 million of funding after 2012.