PROCESSORS

UPDATE -- OctigaBay introduced today a new High Performance Computing (HPC) platform that promises three times greater price/performance over state-of-the-art HPC servers. The OctigaBay 12K combines breakthrough interconnect and management technologies with AMD Opteron 64-bit processors, overcoming major shortcomings that severely limit performance, reliability and usability. These innovations make the OctigaBay 12K ideal for simulation, modeling, searching, sorting, data mining and other processing intensive applications. “OctigaBay has developed an architecture that marries the performance of large SMPs with the economics of cluster solutions,” said John Seminerio, president and CEO of OctigaBay. “The innovations we’re bringing to the market in interconnect, management and reconfigurable computing will not only meet current market needs, but will also change the face of high performance computing.” Interconnect Thirty times faster than Gigabit Ethernet, the OctigaBay 12K high speed interconnect fuses the AMD Opteron 64-bit processors directly to a high performance interconnect fabric through an innovative communications technology, eliminating bottlenecks inherent in today’s solutions. Management Dedicated hardware and software constantly monitor and self heal the system, providing exceptional reliability and availability. Intuitive management software provides total system command and control, which allows administrators to manage up to 12,000 processors as a single system. Reconfigurable Computing Specialized high performance processors deliver application acceleration to speed up computationally intensive portions of HPC applications. With the rapid industry adoption of cluster computing with hundreds to thousands of low cost processors, the limiting factor in scaling performance becomes a system’s ability to feed the processor with data. Traditional high performance computing architectures connect multiple processors through costly shared memory technology (SMP) or I/O buses (cluster systems), which greatly limit scalability and performance. OctigaBay’s Direct Connected Processor (DCP) architecture addresses these limitations by fusing the processor directly to the interconnect fabric, eliminating memory contention and PCI bus bottlenecks. "HPC system designs have gone through an evolution starting with small SMPs in the 4 to 16 processor range, growing to 128 and beyond. Larger SMP designs provide ease-of-use, but become complex and expensive to build. A newer popular approach is clustering many less expensive PC class computers. This approach provides the lowest cost alternative, but becomes more complex to use and manage in larger configurations. There is a competitive race underway to make larger systems easier-to-use and easier-to-manage,” said Earl Joseph, IDC Program vice president. “OctigaBay is addressing these issues with an approach called DCP -- Direct Connected Processor systems, which directly connect larger numbers of AMD processors to the interconnect fabric providing both low cost solutions and ease-to-mange and ease-of-use capabilities." "Balanced system design is critical to application performance," said Mark Seager, assistant department head for Advanced Technologies in the Computing Directorate at Lawrence Livermore National Laboratory. "A well balanced system must address the bottlenecks between memory, processors and interconnect, balancing processing power with high bandwidth and low latency. Innovations in interprocessor communications and process synchronization improve the scalability of our parallel applications solving really demanding stockpile stewardship problems." “Our researchers are constantly pushing the bounds of science and have an insatiable demand for computing resources” said Steve Jones, head of bioinformatics at the BC Genome Sciences Centre and director of bioinformatics for Genome BC. “There are avenues of research that we simply are not able to consider because of performance limitations of today’s solutions. With its reconfigurable computing capabilities, the OctigaBay 12K will give us the potential for an order of magnitude application speedup, opening the door to new areas of research and providing ourselves and our collaborators with a key competitive edge.” About the OctigaBay 12K The revolutionary architectural design of the OctigaBay 12K outperforms existing SMP platforms, while the use of off-the-shelf components dramatically reduces overall cost. Each OctigaBay 12K shelf provides: • 12 AMD Opteron 64-bit x-86 compatible processors for 58 GFLOPS; • HPC-optimized Linux operating system; • 12 specialized communications processors and a 1 Tb/s embedded switching fabric provide up to 8 GB/s to each AMD Opteron with 1 microsecond latency; • Dedicated system management software, hardware and supervisory network; • Active Manager GUI for ease of operation, maintenance and management; and • Six Field Programmable Gate Array (FPGAs) processors for application acceleration. Up to 12 OctigaBay 12K shelves can be assembled in a single rack, providing 691 GFLOPS of processing power. OctigaBay 12K systems scale to over 12,000 processors without requiring any external switches. The OctigaBay 12K system will be featured at SuperComputing 2003 in Phoenix, AZ, November 17 - 19, 2003. Systems will be commercially available early in 2004.