Business Tech

This week's Supercomputing 2009 in Portland, Ore., puts the spotlight on the highest of high-end computing systems.

Intel unveils supercomputer chip, NEC partnership

The chipmaker has disclosed a version of its Xeon processor line optimized for high-performance computing. The company also announces it's teaming up with NEC.
(Posted in Nanotech: The Circuits Blog by Brooke Crothers)
November 16, 2009 1:35 PM PST

Microsoft testing Excel for supercomputers

At the SC09 conference, Microsoft releases an updated version of Windows Server for high-performance computers as well as a compute cluster version of its Excel spreadsheet.
(Posted in Beyond Binary by Ina Fried)
November 16, 2009 9:40 AM PST

Jaguar supercomputer races past Roadrunner in Top500

The Cray XT5 supercomputer known as "Jaguar" finally bests IBM after three tries. The top 10, while still dominated by supercomputers housed in the U.S., had just one newcomer.
(Posted in Circuit Breaker by Erica Ogg)
November 15, 2009 9:00 PM PST

previous coverage

Roadrunner supercomputer maps HIV family tree

Researchers are using IBM's Roadrunner to analyze tens of thousands of genetic sequences from individuals with HIV in the hopes of zeroing in on vaccine target areas.
(Posted in Health Tech by Elizabeth Armstrong Moore)
October 28, 2009 6:25 PM PDT

Supercomputer to use new Nvidia 'Fermi' chip

Oak Ridge National Laboratory announces plans for a new supercomputer that will use Nvidia's next generation GPU architecture.
(Posted in Nanotech: The Circuits Blog by Brooke Crothers)
September 30, 2009 3:05 PM PDT

How Intel's supercomputer almost used HP chips

In the 1990s, Intel seriously considered building the world's fastest supercomputer with a rival's processors, but the Pentium Pro arrived in time after all.
(Posted in Deep Tech by Stephen Shankland)
September 22, 2009 8:09 AM PDT

Intel on Monday disclosed a version of its Xeon processor line optimized for supercomputers and announced a partnership with NEC to develop future supercomputers.

At Supercomputing 2009 in Portland, Ore., Intel unveiled a future version of its "Nehalem-EX" processor optimized for supercomputers. The six-core chip will run at higher speeds than eight-core versions of the Nehalem-EX processors and will offer advantages for supercomputer specific tasks, Intel said in a statement. Intel also refers to supercomputing as high-performance computing, or HPC.

The chip architecture will offer greater memory speeds and capacity and will allow customers to build single computers or "nodes" with up to 256 such processors, according to Intel. This will be available next year, Intel said.

Intel said Monday that four out of every five supercomputers on the Top500 list published Monday are powered by Intel processors.

Intel also announced that it is partnering with Japan's NEC--that country's largest supercomputer vendor--to jointly develop technologies "that will push the boundaries of supercomputing performance," according to a joint statement.

NEC will use the technologies in future supercomputers based on the Intel Xeon processor and other technologies such as AVX (Advanced Vector Extensions), an extension to Intel's x86 instruction set architecture.

AVX will be used with Intel's upcoming Sandy Bridge microarchitecture due in 2011, according to Intel.

"With NEC further innovating on Intel Xeon processor-based systems, Intel is poised to bring Intel Xeon processor performance to an even wider supercomputing audience, " said Richard Dracott, general manager of Intel's High Performance Computing Group, in a statement.

Fumihiko Hisamitsu, general manager of HPC Division at NEC, said: "NEC's substantial experience in the development of vector processing systems...is a natural fit for taking Intel architecture further into new markets."

A vector processor design can perform operations on multiple data elements simultaneously. Intel Xeon chips are good at scalar processing, which handles one data item at a time.

The initial focus of the collaboration will be the development of technology to boost the memory speed and scalability--the latter refers to expanding a system to increase performance or capacity. "Such enhancements are intended to benefit systems targeting not only the very high end of the scientific computing market segment, but also to benefit smaller HPC installations," the two companies said.

NEC will also continue to sell its existing SX vector processor-based products. NEC, for example, currently markets the SX-9 supercomputer.

The Cray XT5 supercomputer known as "Jaguar" has finally clawed its way to the title of fastest computer in the world.

Sitting back at No. 2 on the Top500 list of supercomputers for more than a year, Jaguar overtook IBM's "Roadrunner" according to the twice-yearly list that will be unveiled Tuesday at the SC09 Conference in Portland, Ore.

Jaguar beat out the competition by showing it can process 1.75 petaflop/s, or quadrillions of floating point operations per second, according to the Top500 Linpack benchmark. IBM's Roadrunner was pushed back to No. 2 by posting a processing speed of 1.04 petaflop/s, a dip from the 1.105 petaflop/s it reached in a June 2009 test. The slower performance this time around is apparently due to a repartitioning of the system.

Every six months when the Top500 List is released the threshold to grab a place on it gets higher. The slowest supercomputer (No. 500) on November's list posted a speed of 20 teraflop/s, up from the 17.1 teraflop/s of six months ago. In other words, what is the slowest computer this time around would have been No. 336 in June.

Kraken, another Cray XT5 system, jumped up two places from its former No. 5 position by posting a processing performance speed of 832 teraflop/s. IBM's BlueGene/P, from Forschungszentrum Juelich in Germany, came in at No. 4 with 825.5 teraflop/s. At No. 5 is China's Tianhe-1, the highest ranking ever for a Chinese supercomputer.

The top 10, while still dominated by supercomputers housed in the United States, had just one newcomer. That would be Sandia National Laboratories' "Red Sky," a Sun Blade system that posted a Linpack performance of 423 teraflop/s.

Just as the last time the list was released, the Top500 list is made up mostly of Hewlett-Packard and IBM computers. HP accounted for 210 of this year's 500, and IBM 185. In terms of processors in use, Intel still enjoys the lion's share, with 80 percent. The most popular operating system is Linux, with 90 percent of the Top500.

Here's the Top 10:

• Jaguar, Cray, Oak Ridge National Laboratory (1.75 petaflop/s)

• Roadrunner, IBM, Los Alamos National Laboratory (1.04 petaflop/s)

• Kraken XT5, Cray, National Institute for Computational Sciences (832 teraflop/s)

• JUGENE, IBM, Forschungszentrum Juelich (825.5 teraflop/s)

• Tianhe-1, NUDT, National SuperComputer Center in Tianjin (563.1 teraflop/s)

• Pleiades, SGI, NASA Ames Research Center (544.3 teraflop/s)

• BlueGeneL, IBM, Lawrence Livermore National Laboratory (478.2 teraflop/s)

• BlueGene/P, IBM, Argonne National Laboratory (458.61 teraflop/s)

• Ranger, Sun, Texas Advanced Computing Center (433.20 teraflop/s)

• Red Sky, Sun, Sandia National Laboratories (423.9 teraflop/s)

For the full Top500 List head to the official site.

SAN FRANCISCO--More than a decade ago, Intel ran into an issue trying to deliver what was to be the world's top-ranked supercomputer: it looked possible that its new Pentium Pro processors at the heart of the system might not arrive in time.

As a result, the chipmaker made an unusual move by paying Hewlett-Packard $100,000 to evaluate building the system using its PA-RISC processors in the machine, said Paul Prince, now Dell's chief technology officer for enterprise products but then Intel's system architect for the supercomputer. Called ASCI Red and housed at Sandia National Laboratories, it was designed to be the first supercomputer to cross the threshold of a trillion math calculations per second.

Intel ultimately met that 1-teraflops performance deadline using the Intel chips, HP dropped its PA-RISC line in favor of Intel's Itanium processor line, and the Pentium Pro paved the way for Intel's present powerhouse status in the server market. But the supercomputing division within Intel was phased out, and ASCI Red was its last job, Prince said in an interview here on the eve of the Intel Developer Forum.

The division had enough independence that it could have used another company's chips, but doubtless eyebrows would have been raised had a rival processor design showed up in such a high-profile machine that ultimately used more than 9,000 processors.

It wasn't the only hurdle the Intel group overcame in the design and construction of ASCI Red, which used ordinary processors but plenty of one-off technology including a customized operating system and Intel's own router chips to send data from through the system.

The first version of the router chip had a data integrity problem, and Intel didn't have time to fully validate a fixed version even though the engineers knew what caused the problem, Prince said. However, in a presentation titled "Statistics for the Common Man," Prince convinced Intel management that a variety of worst-case scenario tests could reduce the validation time from more than a dozen weeks to about four to six weeks. He prevailed.

"It worked, and they didn't fire me," Prince said. ASCI Red, developed for the Energy Department's Accelerated Strategic Computing Initiative to simulate nuclear weapons physics in a computer rather than with real-world tests, led the Top500 list of supercomputers from June 1997 until November 2000, when IBM's ASCI White took the top spot.

Meanwhile, in today's world
Naturally Prince now is focused on the best directions for getting Dell servers, storage, and networking gear into customers' hands. And though he's comfortable with nitty-gritty chip details, he said customers these days are gravitating toward higher-level discussions.

"At this point nobody's keeping up with the gigahertz rating of chips," he said, no doubt to the delight of Intel and AMD, who ran into physical limits on clock speed and focused their attention on multiple processing cores and getting more work done in each tick of a chip's clock.

Instead, he said, customers are asking, "How does this fit into my virtual environment? What's my management look like?" Thus, Dell is leading a lot of marketing with virtualization, which lets a single physical computer house many independent operating systems called virtual machines. Dell had expected Microsoft and various Linux players to challenge virtualization expert and EMC subsidiary VMware, but it's withstood the competition so far, he said.

Dell itself has about 6,000 VMware-hosted virtual machines running on about 620 real machines in its own computing infrastructure, but that's only a small fraction of the 12,000 physical servers total the company has. Some physical machines house as many as 20 virtual machines, but for business-critical tasks, Dell puts 10 virtual machines on a physical server, Prince said.

In Dell's analysis, using virtual machines saved $60 million in capital equipment expenses, he said. But virtualization poses problems, too--the virtual equivalent of server sprawl, in which new servers are added to a company's infrastructure faster than administrators can keep up.

"You can deploy new servers in hours instead of weeks. The downside is you crank 'em out, so you have this proliferation of resources," Prince said, and virtual machines don't come with handy tracking technology. "The reason it's hard to get rid of them is it's hard to track them. There's no asset tag. There's no depreciation on a virtual server."

Hardware still matters
Though sales have moved to a higher level, hardware details still matter, Prince said. One he's most excited about is solid-state drives, which use flash memory rather than the spinning platters of conventional hard drives.

Many SSDs today directly replace hard drives, using the same size and SATA or SAS communication protocols to connect to a machine in a way that makes them interchangeable with conventional hard drives. But Prince is more interested in a technology that bypasses that older hard drive technology in favor of a more direct connection over a computer's PCI Express subsystem.

Companies including Fusion-io and Texas Memory Systems supply the technology, and Prince is among those in the server realm who like the idea. "You can get a massive performance upgrade in terms of IOPS," or input-output operations per second.

He's also a believer in a technology called wear leveling, which moves data around the physical storage device so elements don't get overused and therefore effectively worn out. "The life has to be better than enterprise-class drives," he said.

Prince also predicted the eventual triumph of Ethernet over more special-purpose high-speed network fabrics, Fibre Channel and InfiniBand. Fibre Channel will reach 16 gigabits per second, probably won't move beyond 40 gigabits per second, but Ethernet is headed for 40 and 100 gigabits per second today with 400 gigabits and even 1 terabit per second on the horizon, he said.

"Everybody is converging on Ethernet as the high-performance fabric of the future," Prince said.

The National Oceanic and Atmospheric Administration announced this week that it has finally completed a nine-year, $180 million project aimed at installing new supercomputers to aid in more accurately predicting weather. The primary IBM supercomputer is now called Stratus. Its backup is dubbed Cirrus.

The new supercomputers, based on IBM Power 575 Systems, are capable of making 69.7 trillion calculations per second. According to NOAA, the faster the calculation speeds, the greater the chances that meteorologists can rapidly update severe weather forecasts as dangerous weather affects local communities. Billions of bytes are entered into the supercomputers each day to help predict the weather more accurately.

Just how important NOAA's new supercomputers are to our understanding and prediction of weather is easily understated.

Right now, Stratus contains about 20 weather models that predict worldwide weather accurately for about five days. A few decades ago, weather models could forecast weather accurately up to only about two days.

Those 20 weather models rarely change. They analyze conditions such as temperature, humidity, and precipitation to give organizations ranging from the National Weather Service to local meteorologists data on which they can base forecasts.

According to Ben Kyger, director of central operations for the National Center of Environmental Prediction, a division of NOAA, "We analyze weather conditions on grids we lay over maps of the world. In order for meteorologists to accurately predict a hurricane's path, for example, NOAA needs to pinpoint weather conditions in 1-kilometer grids of distance." Right now, those spans "are not even close to that."

How does it work?
In order to improve forecasting, a lot of work needs to be done. Right now, scientists from around the world are analyzing Stratus' weather models to find ways to improve them. When they think that they've come up with an improvement, NOAA analyzes the new models.

If it likes what it sees, NOAA takes it open source. It installs the new model on the Cirrus supercomputer to run in parallel with the approved model on Stratus. Scientists, weather experts, and even you and I can view the new model and inspect it for errors. Errors found are removed or tweaked. If no errors can be found, and the new data enhances weather forecasting, it will be put into operation and replace the existing model that it improved upon.

... Read more

Scientists will collaborate with as-yet-unnamed hardware and software vendors to develop a prototype 100Gbps Ethernet network, which will be used to connect U.S. Department of Energy supercomputer centers.

The aim is to develop a network capable of handling 1Tb (terabit) per second, according to Michael Strayer, head of the Department of Energy's office of advanced scientific computing research.

"This network will serve as a pilot for a future network-wide deployment of 100Gbps Ethernet in research and commercial networks, and represents a major step toward the DOE's vision of a 1Tb--1,000 times faster than 1Gb--network interconnecting DOE Office of Science supercomputer centers," Strayer was quoted as saying in a statement.

The network will be used by scientists to share data and research in such areas as climate-change modeling, and for collaborative projects such as the Large Hadron Collider, the world's largest particle accelerator. Businesses will benefit as 10Gbps and 1Gbps networks will become more affordable, said the statement.

Ethernet networks normally run at either 100Mbps or 1Gbps, while the standard for the fastest is 40Gbps. Scientists working at the Lawrence Berkeley National Laboratory will undertake research into the high-speed Ethernet project, which is called the Advanced Networking Initiative.

The U.S. government has pumped millions of dollars into the project. The Energy Department's ESnet, formally known as the Energy Sciences Network, announced Monday that it had received $62 million in funding.

ESnet, which is run from the Lawrence Berkeley National Laboratory, will put some of the money into new jobs for network and software engineers at Berkeley Lab. However, the bulk of the cash will be used to buy networking equipment and services from providers adjudged to have the necessary infrastructure to support 100Gbps technology.

The funding was allocated under the American Recovery and Reinvestment Act, a stimulus package enacted by the Obama administration, designed to aid U.S. economic recovery during the global financial downturn.

Juniper Networks announced the industry's first 100Gbps Ethernet router interface card in June.

Tom Espiner of ZDNet UK reported from London.

Big Blue's supercomputers are among the greenest in the world.

An IBM supercomputer won first place in a new list ranking the world's most energy-efficient supercomputers.

The June Green500 list, announced June 30 and published by Green500.org, also showed that 18 of the top 20 greenest supercomputers in the world are made by Big Blue.

The group also said that the average efficiency of the supercomputers rose by 10 percent, even as the aggregate power of the machines on the list increased 15 percent.

A key factor in determining a supercomputer's energy efficiency is the number of operations per watt.

Winning the title as most energy-efficient system was an IBM supercomputer based on an IBM BladeCenter QS22 located in Poland at the Interdisciplinary Center for Mathematical and Computational Modeling at the University of Warsaw. The computer produces more than 536 Mflops (millions of floating point operations per second) per watt of energy.

The world's fastest supercomputer, the IBM supercomputer at Los Alamos National Laboratories, came in fourth for energy efficiency, producing over 444 Mflops per watt of energy.

"Modern supercomputers can no longer focus only on raw performance," said David Turek, vice president of deep computing at IBM. "To be commercially viable these systems most also be energy efficient. IBM has a rich history of innovation that has significantly increased energy efficiency of our systems at all levels of the system that are designed to simultaneously reduce data center costs and energy use."

The Green500 group also noted that the No. 5 supercomputer, GRAPE-DR of the National Astronomical Observatory of Japan, is "arguably" the first on its list with more than a million processing elements--in this case, 2.1 million.

Unveiled in 2007, the Green500 list is published two to three times a year by Green500.org. It typically serves as a follow-up to the Top 500 list of worldwide supercomputers announced by Top500.org. In the most recent Top 500 list revealed last month, the Los Alamos supercomputer built by IBM hit a peak performance of 1.105 petaflop/s (quadrillions of floating point operations per second).

Yahoo has signed up three new universities to participate in Internet-scale computing research, the Internet pioneer said Thursday.

The University of California-Berkeley, Cornell University, and the University of Massachusetts-Amherst have joined an effort that already included Carnegie Mellon University, Yahoo said Thursday. The universities get access to a cluster of Yahoo computers called M45 that runs open-source software called Hadoop that can be used to process data rapidly.

Yahoo is a major contributor to Hadoop, a project within the Apache Software Foundation's collection, but Google created the underlying technology through its MapReduce algorithm. MapReduce and Hadoop can be used for tasks such as finding, relatively rapidly, all the Web sites that link to a particular Web site, a task that's essential to the companies' search engines.

Berkeley plans to investigate "societal-scale information" including voting records, polling data, and online news. Amherst plans projects involving the million scanned books in the Internet Archive. Cornell has its eye on biodiversity, socio-economic research, and renewable energy.

The universities also will get access to a research computing research project called Open Cirrus spanning several data centers internationally, Yahoo said. The M45 cluster is part of Open Cirrus, which is run by Yahoo, Hewlett-Packard, Intel, the University of Illinois at Urbana-Champaign, the Infocomm Development Authority in Singapore, the Karlsruhe Institute of Technology in Germany, and the National Science Foundation.

This story has been corrected. See below for details.

IBM plans to announce on Tuesday that it will supply the world's fastest supercomputer to the U.S. Department of Energy in the next few years, according to numerous reports.

Not only will the machine, called Sequoia, be the fastest supercomputer to date, it will blow the current record-holder out of the water. IBM's Roadrunner, located at the U.S. Department of Energy's Los Alamos National Laboratory, was the first system to reach 1.026 petaflops (a petaflop is equal to a quadrillion calculations per second; the "flops" stands for floating point operations per second). But only seven months after the Roadrunner took top honors on a twice-yearly list of the world's fastest supercomputers, IBM is announcing that its successor will outdo it by an order of magnitude. Sequoia will be able to work at a staggering 20 petaflops, the equivalent of the computing power of 2 million laptops according to Reuters.

IBM says it plans to deliver the Sequoia to the Energy Department for use at the Lawrence Livermore National Laboratory. The supercomputer will run simulations to test the soundness of the nation's stockpile of nuclear weaponry, according to the IDG News Service.

With Sequoia, IBM continues to make its supercomputers more energy-efficient as it makes them more powerful. Sequoia will draw 6 megawatts of power. "Though orders of magnitude more powerful than such predecessor systems as the 100 teraflop ASC Purple (4.8 megawatts) and the 590 teraflop BlueGene/L (about 2.5 megawatts), Sequoia will be 160 times more energy efficient than Purple and 17 times more efficient than BlueGene/L (looking at cost per teraflop of computing power)," according to Lawrence Livermore spokesman Don Johnston.

Editor's note: When it was initially published, this story cited inaccurate data from another publication about Sequoia's energy usage. The units used have been corrected, and the story has been updated with more information from Lawrence Livermore.