20-Aug-91 00:34 Thomas.Gross Intel announces iWarp Products From: Thomas.Gross@IUS3.IUS.CS.CMU.EDU NEWS RELEASE Intel Extends Parallel Supercomputing To Real-Time Applications iWarp boards, assemblies and systems provide building blocks for real-time supercomputing in advanced signal and image processing. Beaverton, OR (August 20, 1991)--Intel Supercomputers, a world leader in parallel supercomputing, has begun commercial shipments of the industry's first large-scale parallel systems for real-time, embedded supercomputing applications. Through an innovative parallel communications architecture, Intel's iWarp(tm) products provide both the communication throughput and the computation performance essential for advanced real-time supercomputing. Intel has accepted orders for 30 systems to date, the largest of which is a 256-processor system offering performance of 5 GFLOPS (billions of floating point operations per second). Initial applications include anti-submarine warfare, radar and sonar signal processing, infrared image processing, robotics, numeric control and neural network simulation. iWarp systems are also scheduled for use in the High Definition Display Technology program of the Defense Advanced Research Projects Agency (DARPA). "Historically, the need for real-time supercomputing has been satisfied by special-purpose designs for specific applications," said Les Furnanz, iWarp program manager at Intel Supercomputer Systems Division. "With shrinking defense budgets and increased emphasis on NDI (Non-Developmental Items), iWarp systems are unique as 'off-the-shelf' solution for these real-time supercomputing requirements. Early iWarp customers seem to agree, because they are already using the system for real applications, not just research." The basic building block for iWarp products is the iWarp cell. Consisting of an iWarp microprocessor and local memory chips, each cell is a computer in its own right, with a self-contained computation and communication units that require no additional "glue logic". The iWarp microprocessor features eight communication links and provides an aggregate, sustained processor-to-processor data rate of 320 MBytes/sec. Delivering peak computational performance of 20 MFLOPS single precision and 10 MFLOPS double precision, the architecture is optimized for "fine-grain" computing, typical of real-time signal and image processing applications. The iWarp product family offers three system models, including a complete suite of application development software: + Single-Board Arrays are designed to plug into Sun workstations as application-specific accelerators. Comprised of 4 to 32 cells, they also satisfy the need for inexpensive program development workstations. + Card Cage Assemblies are self-contained processor arrays that support the 19" rack form-factor required for many embedded applications. + iWarp Systems are large-scale multi-computers that expand to 1,024 processors with a peak performance rating of 20 GFLOPS, and support the need for large scale application demonstration vehicles. The four-cell Single Board Array is priced at $39,950 and the 64-cell minimum system is priced at $494,950, and includes host interface and system software. For application development, iWarp systems support a complete programming environment including FORTRAN and C languages with communication extensions, and a full set of Unix programming tools. iWarp also supports two powerful array-level compilers, Apply and Assign, that generate highly efficient parallel-programs. Based on research conducted at Carnegie Mellon University, Apply addresses the needs of image processing applications, and Assign focuses on signal processing requirements. These purpose-built programming tools dramatically reduce the time required to develop code for their respective application domains, while achieving performance that is comparable to hand code. Apply is available immediately and Assign is targeted for Q2, 1992 shipment. In addition, ParaSoft of Pasadena, developer of the Express parallel programming environment, has announced their intention to add iWarp to their line of supported products. Express will benefit iWarp users by providing application portability between iWarp and other parallel systems. "The iWarp product line extends Intel's already established leadership in distributed memory supercomputing to also serve the needs of real-time applications," said Robert Rockwell, general manager of Intel Supercomputers. "With the technical challenge behind us, iWarp is the first of several future products that will benefit from this technology." Intel's iWarp products are the result of a five-year collaboration between Intel and Carnegie Mellon University, jointly funded by Intel and DARPA. Intel was selected to develop the iWarp system in a competitive procurement that included several major U.S. defense contractors. Intel was awarded the development on the basis of cost and technical merit. In cooperation with Carnegie Mellon and DARPA, Intel will conduct a one-day iWarp forum and program review Sept. 11 at the Hyatt Crystal City in Arlington, VA. This conference will focus on application and research results obtained by iWarp users. Intel Supercomputers, a division of Intel Corp., is a leading manufacturer of massively parallel supercomputers, with over 300 systems installations throughout the world. Intel Corp. is an international manufacturer of components, modules and systems. # # # Intel is a registered trademark and iWarp is a trademark of Intel Corporation. Contact: Mike Bernhardt Doug Black Intel Supercomputers Miller Communications Marketing Relations Public Relations 503/629-7835 617/536-0470 Paul Wiley Intel Supercomputers iWarp(tm) Marketing 503/629-6350