By John Abbott

Silicon Graphics Inc has won US Government support to help fund the development of its follow-on to the SV1 supercomputer, the company announced yesterday. SGI will continue with its plans to spin Cray off as a separate, privately-held company, but it now looks as if no other computer firms will be involved as partners. Details of the spin-off are expected to be released shortly, in conjunction with currently unnamed private partners who will provide additional working capital.

SGI says that significant financial development support will come from several government agencies, including the National Security Agency, to help it develop the next generation SV2 vector and scalar processor. George Cotter, chief scientist at the NSA said the systems are absolutely essential to US national security interests. Funding details weren’t revealed, but are thought to be in the tens of millions area for year one, with more to follow for a total of five years. SGI said the development costs for the SV2 would be split on a 50-50 basis, and the added support would enable it to bring the system to market within three years. John R Beau Vrolyk, senior VP of SGI’s Product Group, said that without the investment it would have taken the company from five to seven years to complete the SV2 development.

Information warfare is a very big piece of what is now going on says Vrolyk and the government has to play in that game. It’s now generally realized that it’s impossible for the private sector to develop a new supercomputer from scratch without customer assistance. SGI has already received significant funding from the Department of Energy for its ASCI Blue Mountain machine at Los Alamos Labs, which it claims is currently the most powerful computer in the world. It has also received similar help from NASA’s Ames Research Center for its Aeronautical Engineering Origin 2000 system, said to be the largest single system image machine ever shipped.

The Cray SV2 is intended as the follow-on to the high-end T90 Cray boxes. SGI recently commenced volume shipments of the SV1 line, which replaces the Cray J90, C90 and YMP systems, and which takes performance levels up to 1.2Tflops. The SV2 is expected to shoot performance up to 30TFlops, said Vrolyk. Like the SV1s, the systems will use a mixture of ultra-high-performance vector and high-performance MIPS scalar parallel processors in a single box. A fully configured SV1 symmetrical multiprocessing node has six MSP vector units and eight parallel units for a total of 32 processors: the SV2 will take the number of processors into the hundreds. The front-end machine will run the Linux operating system, while the vector processors will run some derivative of Cray’s own Unicos Unix implementation.

Vrolyk says SGI was misunderstood when it announced that it would spin-off the Cray unit on August 10 (CI No 3,722). People thought we were getting rid of Cray because there was something wrong with it. But we bought Cray many years ago [in 1996, for $750m] while it was losing money, and re-sized and re-directed it so that it can now live on its own as a separate entity. Cray’s market, however is small and specialized, and accounts for only 10% of SGI’s $2.7bn revenue. Under current VP of engineering Steve Oberlin, Cray will continue with the co-development with SGI of a next-generation ccNUMA products, while the Chippewa Falls, Wisconsin and Eagan, Minnesota plants will be responsible for the Cray T90, and the new SV1, SV2 and T3E.