ACADEMIA

MOUNTAIN VIEW, Calif. -- SGI today announced that its SGI Altix 3000 servers and superclusters deliver world-record performance on the next-generation Intel Itanium 2 processor. Preliminary results of 64-bit application tests reveal that the SGI Altix 3000 family running on Madison will once again provide record-shattering performance, price/performance and scalability in a standard Linux OS environment. Based on early industry-standard benchmarks, SGI also expects the combination of SGI Altix 3000 and the Madison processor to deliver system performance over 50% faster than of current record-breaking results on the Itanium 2 processor-based Altix 3000 system. A battery of real-world application benchmark tests confirms the dominant performance and price/performance advantage of the SGI Altix systems, which are up to 130% faster-more than double the speed of the closest competing system. SGI Altix 3000 systems deliver leading performance across the board on popular compute intensive applications from HPC areas such as computational chemistry, biosciences and computational fluid dynamics. Conducted on an SGI Altix system running four 1.5 GHz Madison processors, preliminary internal tests show how Altix easily outpaces the industry's closest competitor, IBM® eServer(tm) pSeries(tm) systems running four IBM® POWER4(tm) processors. Current evidence indicates that these numbers are not expected to vary much when applied to the new POWER4+(tm) processor. SGI intends to issue final performance results following Madison's first shipment, expected this summer. The results will include those for Altix 3000 configurations that uniquely scale up to 64 processors. Leading Performance on Compute Intensive Applications In several different types of jobs using one to four processors, the SGI Altix 3000 system performed 60%, 1.6 times, to 130%, or 2.3 times, faster than the four-processor IBM® p690 system in measuring real-world application performance of Gaussian®, an advanced technical application that allows scientists to predict energies, molecular structures, and vibrational frequencies. (For technical details, visit http://www.gaussian.com ) 1 Tests based on the AMBER® molecular simulation application, administered by the University of California at San Francisco and developed by consortium headquartered at SCRIPPS Research Institute, revealed a 61% improvement over IBM p630.2 In testing the BLAST® (Basic Local Alignment Search Tool) application from the National Center of Biotechnology Information, a suite of tools designed to identify similar protein and DNA sequences within genomic databases, SGI Altix ran 57%, or 2.3 times, faster than the IBM system. In tests of HTC-BLAST, SGI's high-throughput driver program for BLAST, SGI performed more than 72%, or 3.5 times, faster than IBM. 3 Using CD adapco Group's STAR-CD(tm) computational fluid dynamics test suite, SGI Altix 3000 produced a 17.5% performance increase over IBM eServer p690, as well as HP® AlphaServer GS1280. As the Altix system scales to 8, 16, 32 and 64 processors, its performance advantage is likely to increase dramatically, underscoring the SGI Altix family's competitive edge over systems with less balanced architectures. 4 "Intel has done an outstanding job building a microprocessor that raises the bar for all 64-bit applications. We're delighted to see early proof that the SGI Altix 3000 family, with its industry-leading shared memory architecture, will take advantage of the utmost advantage of every possible performance gain of the Madison processor," said Dave Parry, senior vice president and general manager of Products and Platforms, SGI. "These real-world tests show Madison isn't just about enterprise databases. With SGI Altix 3000, Itanium 2 systems will play a significant role in the vastly more challenging high-performance computing realm of 64-bit workloads." Portability and Performance The impressive performance results reported from running Amber and STAR-CD on the Madison processor were achieved by using SGI® Message Passing Toolkit (MPT), the highly tuned SGI message passing interface (MPI) library that provides versions of industry-standard message-passing libraries optimized for SGI systems. These high-performance libraries permit application developers to use standard, portable interfaces for developing applications while obtaining the best possible communications performance on SGI systems for both the IRIX® and Linux operating systems. The MPI implementations in SGI MPT are fully compliant with the current MPI 1.2 specification. Because they are tailored to the SGI® NUMAflex(tm) system, they deliver lower latencies and higher bandwidth than MPICH, a public domain MPI. With MPT using the SGI shared-memory access library (SHMEM), one-sided (put/get) communication and other features in MPT can provide as little as half the send/recv latency and many more times the bandwidth normally achieved with MPICH. Additional information is available at www.sgi.com/software/mpt/ Strong Momentum Building These strong preliminary results build on the swiftly growing popularity of the SGI Altix 3000 family. Since launching the systems in January, SGI has delivered Altix 3000 servers and superclusters powered by thousands of Intel Itanium 2 processors. The new systems, in configurations ranging from four to hundreds of processors, have been installed throughout the world and across all major technical computing markets, including life sciences, physical sciences, oil and gas, manufacturing, and government and defense environments. Inspired by the success of the SGI Altix family, developers to date have moved 43 commercially available high-performance 64-bit manufacturing, science, energy and weather modeling applications to the Linux OS environment, over half of which have certified for the platform. Because SGI made its Linux implementation binary-compatible with standard Linux OS environments, many developers are able to run their code efficiently on SGI Altix 3000 systems without the need to recompile. The result is an easy transition to 64-bit Linux for application developers and users seeking to take advantage of record-setting SGI systems and the open source benefits of Linux. "With the Altix, SGI provides the high performance of IA64 processors in a hardware and software environment which facilitates expansion to large systems and which provides the memory bandwidth and I/O capabilities required for serious production computing," said Mike Frisch, CEO, Gaussian. Balanced Architectures for Record-Shattering Performance For technical computing users, it is crucial to measure the real-life, sustained performance of working systems in the field rather than relying solely on theoretical peak-performance rankings that may not translate into real application speed. In addition to its long history of leading industry-standard performance benchmarks, SGI is committed to delivering balanced computing architectures that allow end users to achieve optimum real-world performance using actual applications. The Altix 3000 family of servers and superclusters combines standard Linux with the Itanium 2 processor family and SGI NUMAflex architecture to enable global shared memory, which is a first for Linux OS-based computing, thus bringing supercomputing capabilities with extraordinary price/performance to Linux users across all markets and applications. Altix can scale up to 64 processors in one node and to hundreds of processors in a supercluster environment. The built-in Altix interconnects deliver outstanding peak and sustained performance in both bandwidth and latency, far exceeding that of traditional interconnects. SGI increases the Altix system's power by offering an advanced high-performance computing software environment that includes the acclaimed SGI® CXFSTM high-performance shared filesystem (available summer 2003), optimized programming and science libraries, and supercomputing data management tools. The SGI Altix 3000 family, fueled by the first high-performance Linux environment capable of scaling to 64 processors in a single node, represents a remarkable leap forward for scientists, engineers, programmers and other users of advanced technical computing systems. Benchmark Sources: 1. Gaussian: Comparing the SGI Altix 3000 running Gaussian 98 rev A11.4 with the IBM (HPC) eServer p690 (at 1.3 GHz) in a frequency calculation problem for Alpha-pinene (C10H16) and a force calculation of Valynomycin (C54H90N6O18). IBM Redbook: Ab Initio Quantum Chemistry on the IBM pSeries 690: A Comparison Between Turbo 1.3 GHz and Turbo 1.1 GHz http://publib-b.boulder.ibm.com/Redbooks.nsf Gaussian 98 and Gaussian 03 are connected systems of programs for performing a variety of semi-empirical and ab-initio molecular orbital (MO) calculations. Gaussian is used to perform fundamental research in chemical, pharmaceutical and material sciences, in academic and commercial settings. http://www.gaussian.com/ 2. AMBER 7, University of California at San Francisco: www.amber.ucsf.edu/amber/amber7.bench4.html AMBER is developed in an active collaboration of David Case at The Scripps Research Institute, the Kollman group at UCSF, Ken Merz at Penn State, Tom Cheatham at the University of Utah, Carlos Simmerling at SUNY-Stony Brook, Tom Darden at NIEHS, Piotr Cieplak at accelrys, and Dave Pearlman at Vertex Pharmaceuticals. Amber was originally developed under the leadership of Peter Kollman, and Version 7 is dedicated to his memory. 3. BLAST, National Center of Biotechnology Information: IBM ~ Performance Technical Report: www.redbooks.ibm.com/redpapers/pdfs/redp0437.pdf Established in 1988 as a national resource for molecular biology information, NCBI creates public databases, conducts research in computational biology, develops software tools for analyzing genome data, and disseminates biomedical information - all for the better understanding of molecular processes affecting human health and disease. http://www.ncbi.nlm.nih.gov/ 4. STAR-CD, CD adapco Group: http://www.cd-adapco.com STAR-CD from Computational Dynamics Limited (www.cd-adapco.com) is one of the CFD technology leaders for fluid flow analysis that provides effective numerical methodologies with the high level of accuracy needed for complex unstructured meshes. www.cd-adapco.com/support/bench/315/mercengine.htm