Super processing power

The high-octane world of Formula One is the bleeding edge of automotive design, with engineering developments such as clutchless manual transmission -- allowing gear changes at the press of a button -- occasionally filtering down to commercial models. But innovation aside, one of the biggest benefits for the companies such as Renault that are involved is the sheen of glamour an association with motorsport adds to its road-going models.

Lacking most of the sparkle and all of the danger, the computing industry's equivalent to Formula One are the monstrous supercomputing projects undertaken mostly in the scientific and technical arena. While these ventures often have worthy aims, and also lead to technical developments that occasionally filter down to enterprises, the real benefits for some vendors again probably comes from the kudos of being associated with a cutting-edge project.

It's no coincidence as chip manufacturer AMD tries to break into the enterprise server market with its Opteron processor it is also heavily involved in two of the most high-profile supercomputing projects in recent months. The first of these projects will see around 2000 Opteron chips utilised by Chinese supercomputer-maker Dawning Information Industry. The Dawning 4000A will be a cluster of machines running a Chinese version of Linux and will aim for around 10 teraflops of computing power.

When launched next June, the Dawning cluster should knock the 2,304-processor Linux NetWorx cluster at the US Lawrence Livermore National Laboratory off its third-place position in the Top500 list of supercomputers. The icing on the cake for AMD is that this 8-teraflop machine just happens to be the fastest on the list using chips from rival Intel.

Dave Everett, European strategic marketing manager for AMD, says while supercomputing projects have benefits in terms of innovation, they are also important in terms of marketing. "I am sure we do get some technical benefits but that's not our primary reason. These projects are very visible and tend to capture the public's imagination," he says.

AMD and Dawning are not the only parties to realise the benefits of some super-powered publicity -- the Chinese government is also behind the project in a big way. The best ranking achieved by a Chinese company in the top500 has been 43, so the Dawning project should help the country's government in its push to be seen as a technical innovator. China's urge to be seen as a techie country is underlined by it recently becoming only the third nation ever to launch a manned space mission.

"We are very excited to see that local Chinese enterprises like Dawning are actually turning the fruit of scientific research into industrial products for the benefit of many. We are happy to see their rapid emergence in the world of science and technology and winning honour for the country," said Shi Dinghuan, secretary-general of the Chinese Ministry of Science and Technology, in a statement.

The second supercomputing project AMD is involved in sees the chipmaker working closer to home supplying the chips for a Cray system, code-named Red Storm, that is being built for the US Department of Energy. The supercomputer will contain around 10,000 Opteron chips and be capable of churning 40 trillion calculations per second (40 teraflops) when it becomes operational in 2004.

Red Storm is part of the DOE's Accelerated Strategic Computing (ASC) initiative to build systems that can, among other tasks, simulate nuclear explosions. But as well as the obvious benefits of cutting down on the need for real nuclear tests -- the project could be a PR coup for the US government in the same way the Dawning system will benefit the Chinese.

The US government had its nose put out of joint when NEC's Earth Simulator project in Japan took the number one spot in the supercomputing chart last year. Part of an initiative sponsored by Japan's Science and Technology Agency -- now called the Ministry of Education, Culture, Sports, Science and Technology -- the system can manage 35 trillion operations per second. But if Red Storm lives up to its promised 40 trillion calculations per second rep, it could push the Japanese system off the top spot -- putting the US firmly back on top of the supercomputing tree.

The US is just as keen as China to be seen as top nation in supercomputing, with the issue even being debated in the highest circles of US politics. Vincent Scarafino, Ford's manager of numerically intensive computing, recently told a hearing before the House Science Committee that: "The Federal government should help with the advancement of high-end processor design and other fundamental components necessary to develop well-balanced, highly capable machines."

In July this year, the US Defense Department also awarded grants totalling more than $146m to Cray, IBM and Sun Microsystems for supercomputing projects as part of its High Productivity Supercomputing programme, which is designed to ensure the US has competitive supercomputers for military, intelligence-gathering and industrial purposes. "The HPCS program will fill a gap in high-end computing that the Department of Defense will experience as it moves from today's high performance computing technology, which dates from the late 1980s, to the future promise of quantum computing," the agency said in a statement.

But while some supercomputing projects have obvious PR benefits to some extent, others offer a source of innovation beyond the scope of the scientific and technical applications they are designed for.

Sun Microsystems grid computing specialist John Barr claims grid computing has its origins in supercomputing projects. "What we are seeing now with grid technology is the next big thing to come out of supercomputing, which is now making the transition into commerce and industry. Grid has slowly been making its way out of academia," he says.

It can be argued that the whole grid concept dates back to the earliest days of computing, but much of the pioneering work around grids came out of early experimental high-speed networks in California and New Mexico. They involved coupling massively parallel and vector supercomputers for computational chemistry, climate modelling and other sciences.

But despite pushing innovations born from the world of high-performance machines, Sun has few actual entries currently in the Top500 list of supercomputing projects -- with its highest current placing at 211. This is especially surprising for a company that prides itself on its pedigree in the high-end server market and its commitment to research and development. "It is partly down the fact that we are not very good at marketing our systems that should be on the list. Having said that there should be a few that are higher up but not a massive number," says Sun's Barr.

He adds that the poor showing in the Top500 is also due to the fact that while supercomputing is a source of innovation it is extremely expensive. "One of the troubles at the high-end of supercomputing is that every organisation buying a big machine thinks there a special case and wants the vendor to drop the prices which is one of the reasons why Sun isn't overly active in some areas. It's not really economic," he explains.

But despite Sun's reluctance to invest in supercomputing -- which also probably has something to do with its relatively poor financial performance lately -- other vendors are still keen to invest. While AMD has made PR waves on a national level recently, rival Intel has an established presence in supercomputing. There are already four Intel Xeon-based machines in the top 20 of the Top500 list and it's likely that Itanium-based supercomputers will be even faster. Some commentators claim that work done in institutions such as the Lawrence Livermore National Laboratory in California to create more powerful Linux on Intel supercomputing systems will lead to the next generation of Itanium-based, general purpose servers.