SGI Altix 3000 - Scalable 64-Bit Production-quality Linux

Next Generation Reconfigurable Compute Technology for Dramatic Application Speedup

SGI® RASC™ RC100 Blade
  • Dual Virtex 4 LX200 FPGAs
  • 80MB QDR SRAM
  • Blade or rack-mountable form factor
  • Dual NUMAlink® 4 ports
  • Seamless direct attach to server's shared memory fabric
  • Datasheet (PDF 140K)

SGI® RASC™ (Reconfigurable Application Specific Computing) technology leverages the power of FPGAs which utilize gate array technology that can be reconfigured by the user for optimal performance on a specific algorithm. Unlike traditional processors, which are serial in nature, FPGAs are inherently parallel, allowing multiple functions to be performed simultaneously. Therefore, users whose applications spend a majority of their run time working on a set of specific algorithms can dramatically reduce application run time by custom configuring the RASC module to accelerate application run-time. This reconfigurable technology is particularly beneficial when running data-intensive applications critical to oil and gas exploration, defense and intelligence, bioinformatics, medical imaging, broadcast media, and other data-dependent industries.

Highly Scalable, Peer-Connected Integration into Shared System Memory for Exceptional FPGA Performance and Unprecedented Levels of Scalability

  • Tightly coupled integration with SGI® NUMAlink® 4 interconnect - the industry's fastest at 6.4GB/sec bandwidth and less than 1 microsecond MPI latency.
  • Direct access to the industry's largest globally shared memory (up to 60TB in SGI® Altix® 4700 server) for faster system level reprogramming of the FPGA.
  • Direct shared memory access enables virtually limitless levels of scalability, with configurations up to 128 NUMAlink connected nodes in a single system image.

Seamless Operation with SGI® Altix® Servers

  • Tightly coupled integration allows RASC technology to seamlessly connect to the NUMAlink fabric as a peer.
  • Direct connection of I/O into the NUMAlink memory fabric for universally accessible memory, compute, and I/O.
  • Processing can be moved from a server or a cluster of servers onto a RASC blade or module without physically changing the memory structure.
  • Enables use of a hybrid implementation, where part of an application can be run on the RASC module, and the other part can be run on a high performance server.

Boosts Price/Performance with Dramatically Decreased Total Cost of Ownership

  • Increases application performance by up to orders of magnitude over conventional servers without the added expense of additional compute nodes or processors.
  • Consumes up to 100 times less power per Gigaflop than a traditional server resulting in savings in power consumption and rack space.

Solution Stack Simplifies Development and Improves Programmer Efficiency

  • An FPGA-aware version of the Gnu Debugger (GDB) built on the current GDB command set provides for simultaneous debugging of both the CPU based application and the FPGA accelerated application.
  • SGI provided Abstraction Layer (RASCAL) that enables serial or parallel FPGA scaling
  • RASC API and core services library provides tools to develop reconfigurable computing elements in a multi-user, multi-processing environment.
  • Support for fully integrated third-party HLL development tools including Mitrionics Mitrion C and Impulse Impulse-C.
  • Synplicity Synplify Pro and Xilinx Synthesis Technology (XST) support for advanced incremental design and modular design methodologies.