GRAPE newletter vol.3 ( Jan 22 2002)

Dear Colleagues:

This is the third issue of a GRAPE newsletter that we intend to send
out every few months, to inform you of new developments.  We will post
this information on our web site as well.

       +----------------------- CONTENTS:-----------------------------+
       | 1) IAU Symposium 208: Astrophysical Supercomputing Using Particles
       | 2) Gordon-Bell Performance Prize awarded to GRAPE-6
       | 3) Grape-6 Reaches 32 Teraflops
       | 4) MUV (Mitaka Underground Vineyard) GRAPE System Upgrade
       | 5) MDM: Massively parallel Grape for Molecular Dynamics
       | 6) Reports from the field: GRAPEs in Tucson
       +--------------------------------------------------------------+

1---------1---------1---------1---------1---------1---------1---------1

 IAU SYMPOSIUM 208: ASTROPHYSICAL SUPERCOMPUTING USING PARTICLES

The above symposium was held in University of Tokyo for July 10-13
(http://grape.c.u-tokyo.ac.jp/iau208/).

36 talks  and 50 posters were presented, covering wide range of
subjects from the formation of moon to the large scale structure of
the universe.

2---------2---------2---------2---------2---------2---------2---------2

 GORDON-BELL PERFORMANCE PRIZE AWARDED TO GRAPE-6

On Nov 16, 2001, the GRAPE-6 was awarded the Gordon Bell prize for
performance.  The calculation submitted was a simulation of a triple
black hole system embedded in the center of a galaxy simulated by
1.1M stars. The sustained performance achieved was 11.55 Tflops
on a 32-board GRAPE-6 system. (see item 2)

3---------3---------3---------3---------3---------3---------3---------3

As of July 8th, a 32-board version of the GRAPE-6, with a peak speed of
32 Tflops, is up and running.  

One processor board houses 32 processor chips, and theoretical peak
speed of one GRAPE-6 chip is currently around 33 Gflops (at a clock
speed of 90 MHz). Thus, single board delivers about 1 Tflops and
the current configuration of 32 boards delivers 32 Tflops.

So far, the best measured performance of actual applications is 11.55
Tflops, for direct simulation of a black hole triple in galactic
center with 1.1 M particles.

4---------4---------4---------4---------4---------4---------4---------4

 MUV (MITAKA UNDERGROUND VINEYARD) GRAPE SYSTEM UPGRADE

National Astronomical Observatory of Japan acquired 8 GRAPE-6 boards
and their host computers (Pentium 4) as an addition to the original MUV
system (16 GRAPE-5 boards and 16 Alpha workstations) in the end of
October. All the G6 boards are connected via the fast network
Myrinet2000 for the parallel use. MUV is currently the world biggest
vineyard, as far as the number of host computers is concerned. The MUV
administrator hopes to start the make public use of the system by
early 2002.

5---------5---------5---------5---------5---------5---------5---------5

 MDM: MASSIVELY PARALLEL GRAPE FOR MOLECULAR DYNAMICS

Researchers in RIKEN (The institute of Physical and Chemical Research,
http://www.riken.go.jp/) have completed the MDM (Molecular Dynamics
Machine), a massively parallel GRAPE system for molecular dynamics
simulations with long-range forces.  The MDM consists of two
special-purpose hardware modules, MDGRAPE-2 and WINE-2, which are
successors to the MD-GRAPE and WINE-1 developed in Tokyo University.
The MDGRAPE-2 handles direct particle-particle interaction. It consists
of 1536 processors chips and has a nominal peak speed of 25 Tflops.
The WINE-2 handles the wavespace part of the long-range interaction
calculation with a direct Ewald method. It consists of 2304 processor
chips with a nominal peak speed of 46 Tflops.

6---------6---------6---------6---------6---------6---------6---------6 

 REPORTS FROM THE FIELD: GRAPES IN TUCSON

 by Matthias Steinmetz (email: matthias@as.arizona.edu)

 web site http://saguaro.as.arizona.edu/~matthias/

GRAPE computers have been used at Steward Observatory since the spring
of 1997 when Matthias Steinmetz moved to Arizona from the MPI fuer
Astrophysik in Garching, Germany. He has been using GRAPE 3Af board
before (since 1993) in Garching and Berkeley. We started with a Sun
Ultra and two GRAPE 3Af boards.  During 1998, this configuration was
extended to a total of 5 GRAPE 3Af board. In 2000 two GRAPE 5 boards,
each connected to a Compaq AXP 1000, arrived in Arizona. The
two-machine unit is currently expanded to a Beowolf cluster featuring
8 nodes each of which featuring a GRAPE 5 board. We expect to complete
this installation in late 2001.

The GRAPE machines are mainly used to study the formation of structure
in the universe, in particular the formation of galaxies and clusters
of galaxies. Our main aim is to understand the origin of galaxy
morphologies (under what conditions does a galaxy evolve into a
elliptical galaxy, under what conditions into a spiral ?) and the
physical processes responsible for scaling laws such as the
Tully-Fisher relation or the fundamental plane relations. On the long
term run we head for simulating the formation of a Milky Way galaxy
with sub-kpc numerical resolution starting from cosmologically
representative initial conditions.

To pursue this science program we mainly run hydrodynamical
simulations using GRAPESPH, a code that combines the Smoothed Particle
Hydrodynamics (SPH) approach to numerical hydrodynamics with a direct
summation N-body integrator for GRAPE, soon to be replaced by a
parallel GRAPE tree integrator written by Vince Eke. GRAPESPH is fully
Lagrangian and highly adaptive in space and time through the use of
individual particle smoothing lengths and timesteps. It is thus
optimally suited to study the formation of highly non-linear systems
such as individual galaxy systems in a cosmological context. The
following physical processes have been incorporated: the self-gravity
of gas, stars, and dark matter, a full 3D hydrodynamical treatment of
the gas, radiative and Compton cooling, and a simple recipe for
transforming gas into stars and for incorporating the feedback of mass
and energy into the gaseous component driven by evolving stars.

0---------0---------0---------0---------0---------0---------0---------0

           Piet Hut and Jun Makino
           (submissions to: piet@astrogrape.org or grape@astrogrape.org)


HOW TO (UN)SUBSCRIBE:  send an email message to grape@astrogrape.org
                       with Subject: (un)subscribe

00--------00--------00--------00--------00--------00--------00--------00