\noindent
{\bf
Connection Machine Model CM-2
}

\noindent
{\bf
Data Parallel Supercomputer (SIMD)
}


\vspace{.1in}
\noindent
{\bf Architecture:}
The Connection Machine architecture assigns
a processor to each element of the program's data.
For example, a 4096 x 4096 array has 16 million elements of data.
Hence it requires 16 million processors.
Virtual processing architecture allows the systems software to subdivide
physical processors into the 
requisite number of virtual processors.  CM-2 systems have a maximum of 65,536
physical processors.  A proprietary chip implements 16 physical processors.
Each processor has 8K bytes of local memory, for a system-wide total of
512 Mbytes.  Information is passed among processors by a very high speed
(3 Gigabits/second) communications path.  
All physical processors may send messages in parallel.  Systems
may be configured with either
32-bit or 64-bit floating-point hardware.

\vspace{.1in}
\noindent
{\bf Configuration:}
CM-2 systems include 4K, 8K, 16K, 32K, and 64K physical processors,
and at least one front-end system (maximum of four frontends in a
configuration).
The front end provides program control and user interaction.
VAX, Symbolics 3600, and Sun front ends are supported.
Large data files are stored on the 10 Gbyte (expandable to 20 Gbyte) Data Vault.
Data Vaults sustain transfer speeds above 20 Mbtyes per sec.  The results of computations may be displayed
on a high-speed graphic display system.

\vspace{.1in}
\noindent
{\bf Software:}
Systems with VAX and/or Sun front ends use the Unix environments of these systems.
Systems with Symbolics front ends use the Lisp environment of these systems.

\vspace{.1in}
\noindent
{\bf Languages:}
System languages are C*, *Lisp, and Fortran.

\vspace{.1in}
\noindent
{\bf Fortran Characteristics:}
Connection Machine Fortran is Fortran 77 with
array extensions from the
ANSI 8x language proposal.
Array extensions allow computations to be carried out on every element of
an array at once.
They include array intrinsics, which are a compact way of specifying global operations
that may change the dimensionality or otherwise alter an array.
Since there is a single program (that operates on all data at once), no
special synchronization commands or debugging techniques are required.

\vspace{.1in}
\noindent
{\bf Applications:}
Applications running currently on CM-2 include molecular dynamics, fluid flow, 3-D elastic wave simulation,
document retrieval, medical imaging, finite element stress analysis,
object recognition, cellular automata, VLSI simulation, and fundamental
physics simulation.

\vspace{.1in}
\noindent
{\bf Performance:}
Peak hardware performance is 31 Gflops.  Rated
performance is 2500 Mflops (the speed
at which the machine
multiplies two large 64-bit matrices).

\vspace{.1in}
\noindent
{\bf Status:}
The first CM-2 system was delivered in September 1987.
System prices range from \$lM to \$7M.

\vspace{.1in}
\noindent
{\bf Contact:}
\begin{flushleft}
Thinking Machine Corp.\\
245 First St.\\
Cambridge, MA  02142-1214\\

\vspace{.1in}
617-876-1111\\
\vspace{.1in}
James Bailey, Director of Marketing\\
\end{flushleft}

.