{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

2 1818 1414 1212 level 3 conceptually dense indexing

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: A set of benchmarks developed… …in the early `90’s by NASA’s Advanced Supercomputing division …to model computations and data access patterns from CFD* codes *CFD = Computational Fluid Dynamics …originally released as paper & pencil benchmarks (v1.x) …then as MPI reference implementations (v2.x) …then versions available in a variety of languages •  Java, OpenMP, HPF (v3.x) •  UPC, Co-Array Fortran, Titanium, ZPL, … (by respective groups) •  Among the most useful benchmark suites in HPC –  –  –  –  well-designed and -maintained good variety of data access patterns, communication requirements open-source well-understood, -used NAS Parallel Benchmarks (NPB) •  8 Benchmarks: –  5 kernels: •  •  •  •  •  EP: embarrassingly parallel MG: multigrid CG: conjugate gradiant FT: Fourier transform IS: integer sort –  3 pseudo-applications •  BT: block transpose •  LU: LU factorization •  SP: pentadiagonal –  Though useful, also domain-specific –  focus on CFD algorithms is good, but restrictive –  other HPC application areas would do well to create similar suites –  Often difficult...
View Full Document

{[ snackBarMessage ]}

Ask a homework question - tutors are online