SGI Packs the Power of Supercomputing Into a Single Blade

With New SGI RASC RC100 Blade, Key Applications See Performance Boosts Surpassing 100x Without Cost of CPU Scale-Out -- Silicon Graphics today unveiled a groundbreaking solution that packs the power of dozens of supercomputer nodes into a single blade by leveraging the inherent parallelism of the industry's most powerful Field-Programmable Gate Array (FPGA) technology. The new SGI RASC RC100 computation blade, built with dual Xilinx Virtex 4 FPGAs, can accelerate the performance of many HPC applications by orders of magnitude over conventional systems at a far lower cost and much smaller footprint. Based on SGI's groundbreaking RASC (Reconfigurable Application-Specific Computing) technology, the new RC100 blade is designed for customers whose applications spend most of their time working on a set of specific routines or algorithms. By accelerating those routines, RASC technology can dramatically improve the performance of the overall application. Designed for use with award-winning SGI Altix servers, SGI RASC RC100 blade can be programmed at the customer's site to accelerate mission-critical, high-performance computing (HPC) applications in oil and gas exploration, defense and intelligence, bioinformatics, medical imaging, and broadcast media. "With certain codes that our researchers run, application performance often boils down to efficiently executing a relatively small set of algorithms again and again," said Warren J. Gross, Assistant Professor, Department of Electrical and Computer Engineering, McGill University in Montreal. "SGI's RASC RC100 Blade solution offers a compelling option for accelerating those very types of algorithms. With SGI RASC, we can also fully leverage the Altix platform's shared-memory architecture, which is crucial for working with large databases." "FPGA technology offers significant computation, I/0 and memory bandwidth advantages over traditional CPU-only solutions," said Ron Renwick, manager, configurable computing, SGI. "HPC users now can take advantage of an FPGA solution built on the award-winning SGI Altix platform that has revolutionized 64-bit Linux computing. With the RC100 blade, the true power of SGI Altix and Xilinx Virtex 4 FPGA technology can be laser-focused on solving a large set of algorithmic-intensive applications-all resulting in tremendous cost benefits to SGI customers." To deliver orders-of-magnitude performance improvements for customers in a wide range of markets, the SGI RC100 blade leverages the SGI Altix system's acclaimed high-bandwidth and shared-memory architecture to create a unique, cost-efficient solution that overcomes the limitations of other FPGA-based products. For instance, the 6.4GB/second SGI NUMAlink(TM) 4 interconnect accelerates memory access, offering more than three times the performance of with PCI and PCI-X bus ports and well outpacing the interconnect technologies supported by companies such as Cray. A robust RASC solution stack incorporates fully integrated third-party high-level language (HLL) development tools, including Handel-C and DK Design Suite from Celoxica and Mitrion-C from Mitrionics, Inc. Along with third-party libraries and tools, SGI provides core services, including a RASC aware version of Gnu Debugger and a RASC API, to ease the process for users who are incorporating RASC RC100 blades as a seamless part of their SGI Altix servers. Available as a blade or add-in module for existing Altix servers SGI RASC technology is available today in two base configurations designed to meet the broadest customer needs. -- The RC100 blade is powered by dual Xilinx Virtex 4 LX200 FPGAs, 80MB QDR SRAM (with up to 20GB DDR2 SDRAM optional starting in June) and dual NUMAlink 4 ports. The RC100 blade is plug-and-play installable on SGI Altix 4000 servers. -- For non-blade architecture systems, a chassis with two blade slots is available that allows users to install RC100 blades on SGI Altix 3700 Bx2 and Altix 350 servers. For pricing and other information, contact SGI or visit its Web site.