High Performance Computing
Evaluation of Early Systems

At ORNL we are constantly pushing the limits of innovation in high-performance computing, bringing the biggest, fastest, and newest systems to bear on problems of national importance.

IBM RS/6000 SP configuration
  • 62 nodes, each with two 200 MHz POWER3 processors and 1 GB memory.
    Planned staged upgrades to nodes with four POWER3+ processors, with 2 GB memory
  • 8 nodes, each with eight 222 MHz POWER3 processors and 8 GB memory.
  • 3.3 TB disk system
Theoretical peak: 128 GF now, 268 GF November 1999, 400 GF January 2000, 1 TF July 2000


Compaq Sierra configuration
  • 16 ES40 nodes, each with four 500 MHz Alpha processors, 2 GB memory, 36 GB local disk and Gigabit Ethernet.
    Planned upgrades to 64 ES40 nodes, each with four 667 MHz Alpha processors.
  • 364 GB fiber channel disk array.
    Planned upgrade to 2 TB fiber channel disk array.
  • Quadrics 200 MB/s interconnect.

Theoretical peak: 64 GF December 1999, 406 GF 1Q 2000


  • Evaluation of reliability, ease of use, scalability, and application performance
HighTORC configuration
  • 64 nodes, each with two 450 Mhz Pentium III processors, 512 MB memory and 8 GB disk. Both Fast Ethernet and Gigabit Ethernet connections.

Theoretical peak: 57 GF


  • Visualization over GRID
  • Tools for HW/SW management of scalable clusters
  • Price/performance of commodity clusters versus commercial alternatives (such as Compaq Sierra and IBM SP)
  • Common Component Architecture (CCA) testbed for Climate and GIST (Genomic Integrated Supercomputing Toolkit)
SGI Origin 2000 configuration
  • 32 R10000 processors (half each 195 MHz and 250 MHz) with 3 GB memory.
  • three-pipe Onyx2 InfiniteReality2 graphics head.

Theoretical peak: 14 GF


  • Regional climate
  • Automotive Crashworthiness
  • Virtual Human
  • Computational Forensics
  • Visualization
SRC configuration
  • Intel Pentium processors, shared memory architecture.
  • Uniform Memory Access (UMA) within unit.
  • User programmable FPGAs for computing.


  • FPGAs as configurable special purpose processors
  • FPGA effectiveness for diverse non-traditional life sciences applications

URL http://www.csm.ornl.gov/SC99/IAO/HDWRwall.html