DESY APE100/Quadrics Configuration
The present configuration at DESY (May 1994) consists of two
Quadrics Q16 (``tube'') and one Q1 (``cube'')
This adds up to
256 + 8 nodes of 50 MFlops each, to a total of 13.2 GFlops
performance peek speed.
Since August 1994, the two Q16 machines for production runs
are located at
DESY-IfH Zeuthen.
General Architecture
The APE100
is a general purpose SIMD computer, based
on a 3-dimensional cubic mesh of nodes with periodic boundary conditions,
each node being connected to its 6 neighbours. The machine has a modular
architecture, the building block being a 2x 2 x 2 cube, while the biggest
version is a 2048-node machine configured as a 8 x 8 x 32 mesh, with
100 GFlops performance peak speed.
Such connection grids are well suited for the simulation of homogeneous
physical systems (including lattice gauge theories and statistical physics).
At a glance, the APE100
provides:
For the modules processing board, crate and maximal configuration:
Characteristic 1 PBoard 1 Crate APE100
-------------------------------------------------------------------
Floating Point 400 MFlops 6.4 GFlops 100 GFlops
Data Memory Size 32 (128) MByte 512 MByte (2 GByte) 8 (32) GByte
RAM to reg rate 400 MByte/sec 6.4 GByte/sec 100 GByte/sec
[the MB in brackets refer to a maximal upgrade of per-node memory,
offered by Quadrics]
APE Future at DESY
By 1994, the present configuration may be enlarged to one Quadrics QH2
(``tower'') with 512 nodes, two Q16 (128 nodes each) and the Q1,
thus reaching 38.8 GFlops nominal speed. After August 1994,
the two Q16's will be moved to
DESY-IfH/Zeuthen in Berlin where
an own parallel computing division is being built up.
Here is more on commercial versions of the Quadrics
.
ms