Mountain View, CA — SGI announced the launch of the SGI 3000 family of systems, which employ a breakthrough in modular design and computer architecture that stands to revolutionize high-performance computing. Available immediately, the SGI Origin 3000 series of servers and SGI Onyx 3000 series of visualization systems promise flexibility, resiliency, overall investment protection, superior performance and scalability.
SGI Origin 3000 series servers and SGI Onyx 3000 series visualization systems utilize the unique SGI NUMAflex modular technology, a “brick”-style system for constructing small to very large systems from a common set of building blocks. The SGI NUMAflex modular system allows users to build the optimum configuration one component at a time and adopt new technologies that map to their specific business or research needs. In contrast, traditional high-performance computers may need to be replaced all at once as often as once a year to keep up with competitive demands and technological changes-a costly and cumbersome process.
SGI Origin 3000 series servers enable “capability computing”-the ability to analyze and solve complex problems that were previously unsolvable. For existing projects or applications, SGI Origin 3000 series servers provide greater precision, quicker results and breakthroughs in price/performance.
SGI Onyx 3000 series visualization systems offer users a unique combination of graphics capability and compute power. This combination allows for visualization of large, complex volumetric data (e.g., brain mapping); allows interactivity and realism (e.g., pilot training simulation); provides bandwidth and image quality for real-time, high-definition special effects (e.g., broadcast); and has the visual accuracy and compute power that enable interactive design (e.g., photo-realistic automotive modeling).
SGI Origin 3000 and SGI Onyx 3000 series systems utilize the SGI IRIX operating system, the world’s premier 64-bit UNIX operating system for high- performance computing, advanced visualization and production supercomputing. SGI IRIX is renowned for its leadership in scalable computation; high-performance data movement, sharing, and management; real-time applications support; and media streaming capabilities. Technical applications that currently run on the SGI 2000 series and Silicon Graphics Onyx2 systems will run on SGI 3000 family systems with as much as twice the previous performance without the need for recompilation. All technical applications currently available on SGI 2000 series and Silicon Graphics Onyx2 systems can run on the new systems
With NUMAflex technology, each drawer-like module in a system has a specific function and can be linked, through the patented SGI high-speed system interconnect, to many other bricks of varying types to create a fully customized configuration. The same bricks, depending on their number or configuration, can be used for a continually expanding range of high-performance computing needs: C-brick (CPU module), P-brick (PCI expansion), D-brick (disk storage), R-brick (system/memory interconnect), I-brick (base I/O module), X-brick (XIO expansion) and G-brick (InfiniteReality graphics). New brick types will be added to the NUMAflex modular offering for specialized configurations (e.g., broadband data streaming) and as new technologies, such as PCI-X and Infiniband, enter the market. The systems can also be deployed in clusters or as large shared-memory systems, depending on users’ needs.
“The scalability, flexibility and performance of these systems are what customers have been asking for,” said Jan Silverman, vice president, Advanced Systems Marketing, SGI. “SGI is proud to be the first to successfully bring modular computing to the industry.”
Customer and analyst reaction to the product launch has been very favorable. Notable SGI 3000 family clients, including the U.S. Army Engineer Research Development Center and NASA/Ames Research Center, have either ordered or already taken delivery. These organizations will use the systems for a variety of needs, ranging from financial analytics to crash-test simulation and aircraft testing. In addition, Sony Computer Entertainment Inc. has selected the SGI Origin 3400 as the broadband server for a next-generation entertainment demonstration at SIGGRAPH 2000.
“One of the key elements when we’re designing a vehicle that is going to fly in the atmosphere and reenter is that it takes a large number of engineers a long period of time-three years or more-to design the vehicle,” said Henry McDonald, director of NASA/Ames Research Center, Mountain View, Calif. “By improving the turnaround time and increasing the number of calculations possible while increasing the fidelity, we reduce the overall development time of the vehicle.”
“The installation of a 512-processor, single system supercomputer from SGI using next-generation SGI Origin 3000 series technology will give government and academic researchers across the country access to the most advanced NUMA shared-memory computing architecture available today,” said Bradley Comes, director of the U.S. Army Engineer Research Development Center’s Major Shared Resource Center (ERDC MSRC), Vicksburg, Miss. The ERDC MSRC is one of four Major Shared Resource Centers established under the Department of Defense High-Performance Computing Modernization Program. Although the system is physically located at the ERDC MSRC, the Arctic Region Supercomputing Center (ARSC) at the University of Alaska in Fairbanks is a partner in the deployment of the new system. Dr. Frank Williams, director of ARSC, added, “We are looking forward to leveraging the combined expertise at the two centers along with the new SGI system to address the large computational requirements the DoD! research and development and test and evaluation communities are demanding.”
“IDC believes that NUMAflex and its current implementation in the form of the SGI Origin 3000 product line should strongly position SGI to regain customer mind share and sales,” said Earl Joseph, Research Director, IDC. “SGI should see strong acceptance of this product in its core technical markets as well as in the markets that service the creative user. Moreover, we see this as a potentially strong product to support emerging Internet workloads given its flexibility, scalability and modular attributes.” For more information on these products, please refer to http://www.sgi.com/origin3000 or http://www.sgi.com/onyx3000 .
Addendum: by Uwe Harms
Munich, GERMANY — In a press meeting SGI announced the long expected successor of the Origin 2000 series, which was named as SN 1. Now the name O3000 was chosen, but additionally a new graphical system with the InfiniteReality was presented. It is based on the Origin 3000, which can be seen as a combination of the best of Origin 2000 and Cray T3E. There will be two lines, a MIPS R12/14000 (400/500 MHz) and IRIX based machine with up to 512 processors. It can be delivered now. With the advent of the Intel Itanium, this processor will be used too with Linux as operating system. The new Origin and Onyx are based on an innovative concept, NUMAflex. Building blocks, called bricks, are developed for different tasks. Thus the machine can be built directly for the specific user’s needs. The first systems have been shipped in America, in Germany the first machine will be installed in August at the Center for High-Performace Computing at the Technical University Dresden.
SGI tries an improved approach with its NUMA (Non-Uniform Memory Access) technology and extends the 128 processor barrier to 512 processors. The next innovation is the modular NUMAflex technology. Today 7 building blocks or bricks are available:
– R-brick router interconnect, it is realized as a high-speed crossbar and replaces the system bus. It connects memory and processors
– C-brick is the processor module and contains 4 MIPS CPUs and local memory up to 8 GByte.. The new crossbar-memory-controller improves the CPU memory bandwidth by 200%.
– I-brick ist the basic I/O module with the basic I/O functionality, system disk, CD-ROM, Ethernet and 4 PCI slots
– P-brick is the PCI extension and provides 12 slots (hot-swappable) with a total I/O bandwidth of more than 3 GByte/s.
– X-brick is the XIO extension and offers 4 XIO slots that support HIPPI, GSN, VME and digital video
– G-brick is the graphics subsystem with the SGI InfiniteReality3 graphic. This allows high-performance visualisation.
– D-brick disk storage, allows a modular integration of JBOD (Just a Bunch of Disks) and RAID and each brick supports up to 12 drives, 16, 36 and 73 GByte.
This approach is flexible and the realization can follow the user’s needs. If he has compute intensive applications, he can buy more C-bricks, if it is I/O bound, I/O bricks are the right solution. All these bricks can be exchanged, when SGI offers new hardware – and thus can be uptodate. Further on, SGI plans to develop more basic blocks for specific applications.
The Origin 3000 MIPS series
SGI offers three models:
Origin 3200 with 2 to 8 MIPS processors, up 16 GB memory, 11.2 GB/s system bandwidth, one I-brick, no R-brick, 18 GB system disk
Origin 3400 with 4 to 32 processors, up to 64 GB memory, 44.8 GB/s system bandwidth, 6-port router (R-brick), one I brick and 18 GB system disk
Origin 3800, 16 to 512 ptrocessors, 1 TB memory, 716 GB/s system bandwidth, 8-port meta router (R-brick), one I- and one P-brick. As the peak performance of the R12000 is 800 (820) MFlop/s, the biggest system sums up to 410 (420) GFlop/s. The high-end system will probably be ranked in the Top500 Linpack-list in the 50s.
The next steps for a bigger system is clustering a solution. Using the integrated Meta Router clusters of thousands of CPUs are realisable with internal clustering. Thus it is possible to realise a single system image with shared memory. If the user wants different partitions for the applications or usage, a single machine image can be defined. The partitions use their own operating system instance. Different operating systems in the different partitions are allowed. The switch between the different modes is realised by software.
The external clustering is done via the NUMAlink interconnct technology.
SGI will build this machine on IA64 – Itanium, when it is available in mass production. The operating system will be Linux. In the beginning this machine will not scale as high as the MIPS series (512 processors).
SGI Onyx 3200 is tuned for small teams, power users, and readily deployable tasks. It scales to up to eight CPUs, two graphics pipelines, 16GB of memory.The SGI Onyx 3000 series is the world’s most p
SGI Onyx 3400, scales from 4 to 32 CPUs and drive up to eight full graphics pipelines and eight simultaneous graphics users.
SGI Onyx 3800, scales from 16 to 512 CPUs and from 1 to 16 graphics pipelines in a single, shared-memory system. For the ultimate in scalability, clusters of SGI Onyx 3800 systems offer thousands of CPUs and hundreds of graphics pipelines.
——- Uwe Harms is a supercomputing consultant and owner of Harms-Supercomputing-Consulting in Munich, Germany.