ASCI White

From TOP500 Supercomputing Sites - Wiki

Jump to: navigation, search

[edit] Description

ASCI White was a supercomputer at the Lawrence Livermore National Laboratory in California.

It was a computer cluster based on IBM's commercial RS/6000 SP computer. 512 of these machines were connected together for ASCI White, with 16 375MHz Power3 processors per node and 8,192 processors in total with 6 terabytes of memory and 160 terabytes of disk storage. The computer weighed 106 tons and consumed 3 MW of electricity with a further 3 MW needed for cooling. It had a theoretical processing speed of 12.3 Tflops. The system ran IBM's AIX operating system.

ASCI White was made up of three individual systems, the 512 node White, the 28 node Ice and the 68 node Frost.

The system was built in Poughkeepsie, New York. Completed in June 2000 it was transported to specially built facilities in California and officially dedicated on August 15, 2001. The system cost $110 million.

It was built as stage three of the Accelerated Strategic Computing Initiative (ASCI) started by the U.S. Department of Energy and the National Nuclear Security Administration to build a simulator to replace live Weapons of Mass Destruction testing following the moratorium on testing started in 1992.

The machine was decommissioned beginning July 27, 2006.

[edit] Information Tables

Vendor IBM
System designer IBM
Model name(s) ASCI White (IBM RS/6000 SP)
Time of manufacture 2000
Number of systems sold -
Resellers -
GENERAL INFORMATION
Class MPP
Type RISC-based distributed-memory multi-processor cluster
Operating system IBM AIX Ver.4.3.3 / 16 CPU per OS
Compilers XL Fortran (Fortran 90), HPF, XL C, C++
Other specific software MPI, PVM
Affiliated technologies -
CPU generation used 64-bit POWER
CPU model 375MHz / 1.5Gflops Power3-II
DRAM type -
FPGA n/a
System coolingAir
Floor space 10000 square feet for system / 106 tons
NETWORK
Node Interconnect -
Network switch /
Size / Quantity
4 Colony switches (8 frames) / 4096 links
Network topology -
Management network -
BUILDING BLOCKS
Block name --> NodeSystem
Structurebasic block512 nodes
Number of CPU 16 8192
Memory size (max) 4-32GB6.2 TB
Other equipment - 160 TB disk storage
Power consumption - 3MW + 3MW for cooling
Theor.performance 24Gflops12.3 TFlops
BENCHMARKS
HPL Benchmark: CPUs /
Rmax / efficiency / Date
- 8192 CPU / 7304GF / 59.4% / 2000
Application performance - -

[edit] External Resourses

Personal tools