NAS Parallel Benchmarks (NPB)

What is NPB?

The NAS Parallel Benchmarks are a set of benchmarks used to evaluate the performance of parallel supercomputers. They are developed and maintained by NASA Advanced Supercomputing Division.
These benchmarks are mostly based on computational fluid dynamics (CFD) applications and are cross-platform.

Where may I find it?

You may download the benchmarks directly from the NPB website. A C with OpenMP version can be found here.

Are there different flavors?

All benchmarks in the set have some differences. There are different versions of the NPB, from 1 to 3. In the NPB website, you may find some versions with OpenMP, Fortran, Java, High Performance Fortran, MPI, etc. In the Omni OpenMP Project Website there is a C with OpenMP version available for download.
The newer versions of the benchmarks in the NPB are free, but NAS solicits the results obtained to incorporate them in a database.

Which one is going to be use here?

We are currently looking into the C implementation with OpenMP. We plan to base our implementations with CUDA in this implementations.
The Omni OpenMP code is based on the NPB version 2.3 using Fortran and OpenMP. There are 8 benchmarks in this version:
  1. BT – Block Tridiagonal
  2. CG – Conjugate Gradient
  3. EP – Embarrassingly Parallel
  4. FT – Fast Fourier Transform
  5. IS – Integer Sort
  6. LU – Lower-Upper Symmetric Gauss-Seidel
  7. MG - MultiGrid
  8. SP – Scalar Pentadiagonal
Some guidelines about building and running the benchmarks on Linux can be found here.

Last edited Apr 17, 2009 at 8:48 PM by Pilla, version 4

Comments

maria_my Jun 16, 2010 at 12:08 AM 
cg , ft and ep kernals of nas bechmarks are good but i want mg kernal implemented in cuda. does anyone have it