SGI launch makes new bid for research community

Supercomputer maker SGI is trying to extend its lifespan in the high-performance computing (HPC) market with the launch of the Onyx4 UltraVision, an upgraded flagship system it hopes will appeal to life sciences and other researchers keen to get ever more detail in the graphical representation of their data.

The company has enjoyed a strong position in the HPC market for years, known particularly for the strength of its systems' graphics capabilities. Combined with solid computing performance, the company's systems have been popular in molecular modelling, visual design, 3D simulation and other environments involving representation of massive, complex data sets.

Like every other supplier of HPC systems, SGI has come under pressure from smaller, nimbler makers building large computers from small industry-standard components such as conventional Intel processors. IBM has enjoyed particular success, beating out SGI and its ilk in deals to supply systems to scientific research computing partnerships in Melbourne and Adelaide.

SGI has been popular for modelling fluid dynamics, molecular structure, 3D models, and other mathematically-based work that can be represented visually. Holden uses SGI systems to design new automobiles down to the individual bolt, for example, while the Defence Science and Technology Organisation will purchase Australia's first Onyx4 to aid in flight-related research. But Intel-based systems continue to grow their market share, doubling their number in www.top500.org's most recent list of the world's top 500 supercomputers.

In that light, the launch of SGI's Onyx4, as well as its new Tenzo desktop workstations, is clearly the company's stab at survival. Such detailed questions require massive amounts of computing power, and SGI's new flagship systems deliver it in spades. Onyx4 scales from two to 32 ATI Technologies graphics processors, supporting compute performance that SGI puts at eight times that of previous generations.

Whereas Intel-based servers from IBM, HP, Unisys and others allocate individual memory segments to each processor, SGI's design is built around a single pool of up to 1 terabyte of memory -- allowing it to render extremely fine detail when visualising complex computing problems. Entry-level Onyx4 systems deliver 20 million pixels, while high-end systems push this over 100 megapixels -- enough to power fully immersive computing environments.

"Designers are aiming for more fidelity in their models, so they can get insight more quickly," says Bill Trestrail, managing director of SGI Australia. "Now, automotive manufacturers are not only wanting to model their cars [in crash simulations] but they also want to get into the human body to find out, for example, what's the damage on capillaries in human lungs."

Pressure from Intel-based competitors is evident in the new systems' pricing: Tenzo workstations are 20 per cent cheaper than SGI's previous Octane2 desktops, while prices for the Onyx4 start at around US$45,000 -- one-fifth the price tag on SGI's previous Onyx InfiniteReality. SGI is also trying to broaden the accessibility of its systems with the introduction of Visual Area Network, a new technology that SGI says will let users who aren't directly attached to the system, access a scaled-down version of its applications to collaborate from notebooks, handheld PCs, and other devices.

Trestrail believes the new system's specifications and competitive price will make it a compelling purchase for funding-starved researchers, for whom price can be a major impediment to supercomputer access. At its old price range, SGI's systems were simply too big and expensive to be widely deployed, but Onyx4 takes a building-block approach that should let the systems be expanded to suit users' changing needs.

"We're moving a lot more to embrace the economies of scale that come from industry-standard servers," Trestrail said. The goal "is to bring pricing down and make it more granular so people can get into them, then get performance increases later by adding processors and graphics in small increments."

The big question? Whether that's enough to convince Australia's research community not to jump aboard the IBM bandwagon. The answer remains to be seen.