[Future Technology Research Index] [SGI Tech/Advice Index] [Nintendo64 Tech Info Index]


[WhatsNew] [P.I.] [Indigo] [Indy] [O2] [Indigo2] [Crimson] [Challenge] [Onyx] [Octane] [Origin] [Onyx2]

Ian's SGI Depot: FOR SALE! SGI Systems, Parts, Spares and Upgrades

(check my current auctions!)
[This is all very out of date now of course, but it's interesting to see what was originally planned - Ian]

Thursday November 14 10:22 AM EDT

Cray Research Shows Path To Unified High-Performance Architecture

Company's Parallel Vector, MPP, and S2MP Architectures Converge to Enable Unprecedented 'Computational Insight'

EAGAN, Minn., Nov. 14 /PRNewswire/ -- Next week at the Supercomputing 96 Conference and Exhibition in Pittsburgh, Cray Research, the supercomputing subsidiary of Silicon Graphics Inc will further outline its plans to develop a unified high-performance architecture for supercomputing. The unified architecture will meld in a single design the most effective attributes of distributed memory scalable parallel (also known as MPP), parallel vector processing and S2MP (Scalable Shared-memory MultiProcessing) -- the three principal high-performance architectures currently provided by Cray Research to meet the needs of the HPC market.

Robert H. Ewald, president of Cray Research and senior vice president of Silicon Graphics, said the unified architecture will integrate leading-edge simulation, visualization and data management capabilities, as well as culminate in the world's first binary compatible desktop-to-teraflops product line to provide a means of solving the world's most demanding scientific and engineering problems.

"Our customers are already working on push-the-envelope' applications that will require interactive simulations steered' through very high resolution graphics and involving the management of huge volumes of data," said Ewald. As examples, he cited applications such as virtual prototyping and design, virtual manufacturing and process design, real-time reservoir modeling and ensemble weather forecasting.

"As has been the practice at Cray Research and Silicon Graphics, we will collaborate with our customers to achieve the breakthroughs required to meet emerging market needs," said Ewald. "Our merger with Silicon Graphics enables Cray Research to focus completely and intently on high- end supercomputing while the combined strengths of Cray Research and Silicon Graphics provide the critical mass of expertise required to deliver an ideal high-performance platform -- one that will enable customers to achieve new levels of scientific, engineering and business insight through interactive, integrated simulation and visualization."

Category Leaders

In the high-performance computing marketplace, customers can choose from Cray's broad range of products covering a full spectrum of architectures based on the best fit to application.

"Each of our systems today are the category leaders in their class of computing," said Ewald. "For instance, our recently introduced CRAY T3E-900(TM) Series systems are optimized for distributed memory, parallel applications that can take advantage of the systems' ability to scale to hundreds and up to thousands of processors. Our recently introduced CRAY Origin2000(TM) supercomputers offer a solution for a broad range of mainstream, third-party, technical, commercial and visualization applications with scalability up to 128 MIPS(R) RISC processors. Meanwhile, the CRAY T90(TM) and CRAY J90(TM) parallel vector supercomputers remain the market-leading supercomputers for vectorized third-party applications," he said.

The challenge for Cray Research and Silicon Graphics is to deliver a platform that offers the distinct benefits of each architecture across the broadest range of customer applications. In consultation with customers worldwide, Cray Research has identified six attributes that define this ideal supercomputing platform. They include:

-- Leading price/performance resulting in an attractive return on investment for customers

-- Near linear scalability offering low entry prices and "pay as you go" pricing, as well as the ability to achieve extreme levels of performance

-- A shared memory architecture offering ease of programming and a broad applications catalog

-- Integrated visualization and data management resulting in intuitive system use

-- Cray-class single processor performance, resulting in sustained gigaflops-level performance

Cray also said in addition to the above list, driving parallel applications is key to its plan.

Milestones on the Hardware Roadmap

A unified architecture addressing these requirements will be introduced in phases over the next several years with new generations of product being offered every two and one-half to three years. The journey, however, has already begun with the launch of the S2MP architecture this fall, Ewald said.

"The S2MP architecture introduced with the Origin family of products, including CRAY Origin2000 supercomputers, lays the architectural foundation for our unified product line," he said. In its initial release, S2MP provides industry-leading price-performance with a fast MIPS RISC processor; high bandwidth-low latency interconnect; an innovative cache-coherent, distributed shared-memory architecture; the ability to integrate compute and graphics capabilities in a single system; and scalability from two to over 100 processors in a single system software environment.

S2MP with Extreme Scalability

The next-generation S2MP product (code-named "scalable node one" or SN1) will provide scalability to thousands of processors as the extreme scalability now available only in CRAY T3E systems is incorporated into SN1.

"You will see us merge the extreme scalability technologies of the CRAY T3E -- such as interconnect, scalable operating system capabilities, applications software and I/O -- with the general purpose, globally cache coherent shared-memory technology of the S2MP architecture. In addition, SN1 will achieve a performance improvement resulting from a step-up in the speed of the binary-compatible individual processors," Ewald said.

MIPS and FLOPS Merge

The third generation S2MP product brings together parallel vector and SN1 in a single architecture. "SN2 will be the first highly scalable system that combines the price/performance benefits of RISC processors with our advanced vector processing technology," said Ewald. "In meeting this challenge, the combined strengths of Cray Research and Silicon Graphics will prove especially valuable. Together, Cray and Silicon Graphics bring to this task distinctive expertise in uniprocessor design, including vector processing, and a microprocessor design in MIPS Technologies dedicated to high-performance solutions. In short, our combined company has the resources and, equally important, strong business and economic incentives, to develop the SN2 processor that will be critical enabling technology for our unified platform," said Ewald.

This third-generation implementation of the S2MP architecture will be scalable to thousands of MIPS processors. "Offering tightly integrated computation, visualization and data management capabilities, SN2 will be the successor to all current and panned Cray Research and Silicon Graphics high-end product families," said Ewald.

Software Environment Will Also Converge

As high-performance hardware architectures migrate to a unified platform, so too will the software environment, which will converge on a common, scalable operating system architecture. Working together, Cray Research and Silicon Graphics intend to meld complementary capabilities from UNICOS and IRIX operating systems to create a "best of both worlds" environment that will protect customers' software investments, mitigate migration issues, ensure availability of a broad range of applications and provide an easy-to-use programming model.

"Our goal is to ensure most customer applications running on Cray Research systems today will port to current and future generations of the S2MP architecture without source code modifications," said Ewald.

The S2MP architecture supports scalable shared-memory parallel programming models and explicit distributed-memory message passing models.

"This flexibility will ease the development of new parallel applications and simplify the porting of existing ones," said Ewald, adding that Cray Research and Silicon Graphics will also continue to work closely with customers and independent software vendors to accelerate the availability of parallel applications, drive the parallelization and optimization of key codes, combine visualization and data management with high-performance simulation and computation.

Getting There from Here: Multiple Transition Paths

As the evolution to a unified high-performance architecture unfolds, Cray Research and Silicon Graphics will provide customers with multiple migration paths.

"Our intent is to make the transition to the new architecture graceful and transparent for all, and most of our customers will see only one transition," he said.

Cray Research's recent introduction of the CRAY T3E-900 Series, hand-in-hand with the commitment to additional enhancements, is representative of how the transition will be handled, according to Ewald.

"Cray T3E-900 Series systems will merge with the next generation of CRAY Origin2000 supercomputers in the post-1998 timeframe as SN1 becomes a fully supported product. Nevertheless, we are planning another enhancement to CRAY T3E performance beyond the CRAY T3E-900 system to provide an upgrade path for customers who do not plan to replace their CRAY T3E systems until after 1998," said Ewald.

Similarly, customers of Cray's parallel vector lines, the CRAY J90 and CRAY T90 systems, will be able to take advantage of planned upgrades to these systems or migrate to SN1 or SN2 implementations of the unified architecture over the next several years. Ewald added that, in terms of single processor performance, CRAY T90 Series systems will remain the fastest systems offered by Cray Research and Silicon Graphics through the end of the decade. "Work continues on a follow-on product to the CRAY T90; the successor of that product will be the SN2 product, which will become available some time after the year 2000," said Ewald.

"The development of our unified high-performance architecture will be the ultimate demonstration of the wisdom of combining Cray Research and Silicon Graphics," said Ewald. "Together, we'll get the job done better and more quickly. The progress we've already made with the introduction of the S2MP architecture attests to how committed we are to moving forward aggressively on our desktop-to-teraflops vision. No other organization is better equipped to fulfill the promise that vision holds."

The discussion in this news release of new products and technologies contains forward-looking statements that involve risks and uncertainties, including the timely release to manufacturing, the availability of components from suppliers, the impact of competitive products and pricing, and the other risks detailed from time to time in the company's SEC reports, including the report Form 10-K for the year ended June 30, 1996. Actual results may vary materially.

Cray Research (Eagan, Minn.) is the supercomputing subsidiary of Silicon Graphics, Inc. Cray Research provides the leading supercomputing tools and services to help solve customers' most challenging problems.

Trademarks: Silicon Graphics and MIPS are registered trademarks and S2MP is a trademark of Silicon Graphics, Inc. Cray is a registered trademark and CRAY T3E, CRAY T3E-900, CRAY T90, CRAY J90, and CRAY Origin2000 are trademarks of Cray Research, Inc. SOURCE Cray Research, Inc.


Ian's SGI Depot: FOR SALE! SGI Systems, Parts, Spares and Upgrades

(check my current auctions!)
[WhatsNew] [P.I.] [Indigo] [Indy] [O2] [Indigo2] [Crimson] [Challenge] [Onyx] [Octane] [Origin] [Onyx2]
[Future Technology Research Index] [SGI Tech/Advice Index] [Nintendo64 Tech Info Index]