Newsgroups: comp.parallel.mpi
From: Bill Saphir <wcs@best.com>
Reply-To: wcs@best.com
Subject: NAS Parallel Benchmarks 2:  results now available
Organization: Best Internet Communications
Date: Mon, 26 Aug 1996 21:11:20 -0700
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <32227566.514@best.com>

Results from the NAS Parallel Benchmarks 2 (NPB 2) are now available
on the World Wide Web at:

           http://www.nas.nasa.gov/NAS/NPB/

This page includes pointers to NPB 2 source code, NAS Technical Report
95-020 describing the NPB 2, NAS Technical Report 96-010 describing the
first NPB 2 results, raw NPB 2 result data, and many graphs of that
data.

============================================
From the introduction to the NPB 2 results report:

The NAS Parallel Benchmarks (NPBs) are a widely-recognized suite of
benchmarks originally designed to compare the performance of highly
parallel computers with that of traditional supercomputers. The NPBs
are specified algorithmically, and are implemented mainly by computer
vendors, using techniques and optimizations appropriate to their
specific computers. Performance results based on these optimized
proprietary implementations are submitted to NAS by vendors and are
reported in a periodic NAS Technical Report. The vendor-optimized NPB
implementations will be referred to as NPB 1.


In late 1995, NAS announced NPB 2, a set of specific NPB
implementations, based on Fortran 77 and MPI. NPB 2 implementations
are intended to be run with little or no tuning, in contrast to NPB 1
implementations, which have been highly optimized by vendors for
specific architectures. NPB 2 implementations are designed for
computers with hierarchical cache-based memories.

Because they have not been optimized by vendors, NPB 2 implementations
approximate the performance a typical user can expect for a portable
parallel program on a distributed memory parallel computer. Together
the results presented here provide a well-calibrated comparison of the
real-world performance of several parallel computers.  Furthermore, by
comparing NPB 2 results to NPB 1 results, we can draw conclusions
about how easy it is to obtain high performance on these systems.  NPB
2 results complement, rather than replace, NPB 1 results.

======================

NAS Parallel Benchmark Group
npb@nas.nasa.gov

