|The Leading Source for Global News and Information Covering the Ecosystem of High Productivity Computing / November 9, 2007|
Unveils RC200 blade to bring FPGAs to Xeon-class SGI system
SUNNYVALE, Calif., Nov. 8 -- SGI today announced it built the world's largest Field Programmable Gate Array (FPGA) supercomputer configuration, then ran a broadly used bioinformatics application more than 900 times faster than the same application would run on a traditional cluster.
SGI's reconfigurable supercomputer featured 70 FPGAs, more than any single system built to date. SGI's FPGA supercomputer accelerated the performance of a complex BLAST-n query by more than 900 times, completing in less than 33 minutes what took a 68-node Opteron-based cluster approximately three weeks to finish(1). The application matched 20 nucleotide base pairs against 600,000 queries.
SGI configured the system using only off-the-shelf components, including its SGI RASC (Reconfigurable Application Specific Computing) appliance for bioinformatics -- Featuring Mitrion-Accelerated BLAST-n. No hardware or software was modified for the test(2).
Bill Mannel, SGI's director of marketing for servers, compared the SGI RASC system to earlier FPGA systems of similar (but smaller) size. "Previously FPGA supercomputers have been custom-built at very high cost," said Mannel. "The SGI RASC system, in contrast, was built with off-the-shelf components in a short period of time and at less than half the cost of the largest of those custom supercomputers. This represents how SGI is bringing its core capabilities in the high-performance computing industry into the reconfigurable compute space."
Many SGI customers have achieved significant performance improvements with SGI RASC deployments incorporating many fewer than 70 FPGAs. Already in its fourth generation, SGI RASC technology has boosted the productivity of data-intensive applications in industries such as oil and gas exploration, defense and intelligence, bioinformatics, medical imaging, and broadcast media.
FPGAs Go Mainstream with New RC200 Blade
To bring the benefits of FPGAs to more users, SGI today unveiled the new SGI RC200 blade. The new blade is the first to bring SGI RASC technology to SGI Altix XE and SGI Altix ICE clusters and blade servers, both of which are based on Intel Xeon processors. Now organizations with applications running on x86-architecture platforms can incorporate SGI RASC technology in their computing systems. SGI developed the RC200 blade with XtremeData, Inc.
"SGI RASC solutions are designed to bring the benefits of FPGAs to more customers, and the RC200 blade is the next important step in that effort," said Bill Brown, server product marketing manager, SGI. "With on-site integration provided by SGI Professional Services, this new blade can improve the performance of their Xeon-class clusters and blade servers."
SGI RASC technology combines advanced FPGAs with software developed by SGI and its partners to reduce the programming bottlenecks that have kept FPGA solutions from more widespread adoption within the high-performance computing (HPC) marketplace. Unlike traditional processors, FPGAs allow multiple functions to be performed simultaneously, enabling users whose applications spend a majority of their run time working on a set of specific algorithms to reduce application run time by custom-configuring their RASC module.
Each RC200 blade combines multiple Altera high-performance Stratix III FPGAs to create the most powerful FPGA module available. It also uses the Intel QuickAssist technology accelerator abstraction layer to provide a connection to the front side bus of the Xeon processor.
"SGI has been a pioneer in driving user-friendly implementations of FPGA technology in the HPC market, and we're excited to work with them on the RC200," said Geno Valente, vice president, sales and marketing, XtremeData, Inc. "We hope to make the benefits of FPGA-based computing, including increased performance, power savings and space efficiencies, available to a broader set of users."
For more information on the SGI RC200 blade or other SGI RASC solutions, visit http://www.sgi.com/products/rasc/.
(1) Results compared to industry-standard Opteron processor-based system measured in internal tests. Running a released version of BLAST, SGI used a test case from Affymetrix comparing approximately 600,000 queries with a query size of 25 base pairs against the Human Unigene and Human ReSeq databases, which is representative of current top-end research in the pharmaceutical industry. Total execution time on a traditional 68-node Opteron-based cluster would be approximately three weeks. On the SGI reconfigurable supercomputer, benchmark input data was split in 169 jobs, which were run in groups of 70, 70 and 29 FPGAs. Total wall clock time for the run was 32m:29.183s, representing a 916X speedup over the 68-node traditional cluster.
(2) The tested SGI configuration consisted of 35 dual-FPGA SGI RC100 RASC blades, a 64-processor SGI Altix 4700 with 256GB of globally addressable memory, and standard SUSE Linux Enterprise Server 10 SP1 (kernel version 22.214.171.124-0.12) running an unmodified release of RASCAL, SGI's RASC Abstraction Layer.
SGI (NASDAQ: SGIC) is a leader in high-performance computing. SGI delivers a broad range of high-performance server and storage solutions along with industry-leading professional services and support that enable its customers to overcome the challenges of complex data-intensive workflows and accelerate breakthrough discoveries, innovation and information transformation. SGI helps customers solve significant challenges whether it's enhancing the quality of life through drug research, designing and manufacturing safer and more efficient cars and airplanes, studying global climate change, providing technologies for homeland security and defense, or helping enterprises manage large data. With offices worldwide, the company is headquartered in Sunnyvale, Calif., and can be found on the Web at www.sgi.com.