Stuttgart, Germany — Silicon Graphics, Inc. unveiled the CRAY SV1 series, a vector supercomputer product line with innovations that create performance, and price/performance up to eight times better than current vector systems. The new product, slated for availability starting in August 1998 from U.S. list prices of $500,000, completes an important milestone on the company’s product roadmap announced in April, officials said.
Among the CRAY SV1-series features company officials disclosed at the opening of the worldwide Cray User Group meeting held here this week:
Processors with theoretical peak performance of four gigaflops, or four billion calculations per second — twice as fast as the company’s fastest processors today; single-cabinet nodes with peak performance up to 32 gigaflops in an easy-to-administer, reliable symmetric multiprocessing (SMP) architecture; and a powerful suite of clustering tools that tightly couples nodes using message-passing programming models to produce systems with up to one teraflop (trillion calculations per second) of peak performance and one terabyte (trillion bytes) of memory.
The world’s first vector cache memory boosts actual processor performance by dramatically increasing effective memory bandwidth.
A mix of 4-gigaflop and 1-gigaflop processors in each system efficiently handles varied problems and workloads.
The world’s first adjustable-size vector processors — for added efficiency, each 4-gigaflop processor can be run as four 1-gigaflop processors.
The company’s robust, fourth-generation CMOS supercomputer architecture for unrivaled price/performance.
The Cray library of more than 500 optimized vector supercomputer applications — by far the world’s largest — and tenth-generation, Year 2000-compliant UNICOS* (UNIX*) operating system make the CRAY SV1 series much more widely usable than competing products.
“The CRAY SV1 series is a breakthrough product,” said Silicon Graphics Chairman and CEO Rick Belluzzo. “For the vast majority of vector customers, it will deliver more real processing power for the money than anything else on the planet, so customers can handle more complex problems, handle them faster, and try out more solutions in the available timeframe.”
“The CRAY SV1 series is also a major milestone for our product roadmap, which will extend our supercomputing leadership into the next century with products that are dramatically more powerful, dramatically more cost-effective for us and our customers, and increasingly unified with Silicon Graphics desktop and server products,” Belluzzo said.
According to Beau Vrolyk, senior vice president, Server and Supercomputer Business Unit, the CRAY SV1 series is aimed primarily at established vector supercomputing markets, where Silicon Graphics has a commanding leadership position. These include government; science and education; manufacturing, especially automotive and aerospace; and the chemical-pharmaceutical sector.
“The CRAY SV1 product line provides an attractive, compatible upgrade path for our installed base of more than 400 CRAY J90*-class systems — in fact, CRAY J90 systems can be field-upgraded with CRAY SV1 processors. CRAY SV1 systems are also compatible with our CRAY C90 and CRAY Y-MP prior-generation high-end systems. The CRAY SV1 series allows these customers to run their current software applications, in most cases with no modification, and presents a compelling economic argument. It also puts them on our exciting roadmap into the future,” Vrolyk added.
“The CRAY SV1 series will also cost-effectively run most high-end supercomputing workloads, thanks to its powerful processors, scalability, and innovative vector caching to boost memory bandwidth,” he said. “Our ultra-high-bandwidth CRAY T90 line can handle the rest.”
Vrolyk expects the CRAY SV1 series to capture new customers. Stating that affordable supercomputing technology is what the market is looking for, he said a substantial percentage of CRAY J90 customers were new to the company. “The CRAY SV1 series redefines the market for powerful, affordable vector supercomputer systems.”
The CRAY SV1 series runs popular programming models and the Silicon Graphics CF90 Fortran compiler, and uses the company’s high-bandwidth GigaRing I/O (input/output) channel.
The Silicon Graphics product roadmap announced in April calls for both the CRAY SV1 series, or “scalable vector first-generation” series, and the existing CRAY T90 vector product line to be succeeded by a second-generation scalable vector series with peak performance of “tens of teraflops,” officials said.