Keywords multiscalar computers, parallel architectures, supercomputers
Start Date: 01 June 1992 / Duration: 36 months / Status: ongoing
[ contact / participants ]
development of a prototype of the most powerful multiprocessor node compatible with existing systems in the chosen application domain
The multiprocessor node will appear to the user as if it were a monoprocessor. It will have a clock speed of a few nanoseconds, and be optimised to have a very high sustained/peak performance ratio.
In parallel with this work, the design and simulation of a multinode system along with the corresponding software extensions is being undertaken.
The architecture is proprietary and will support decoupled operations, hardware-assisted loop pipelining and a very large hierarchical memory. Highly optimised compilers will be developed to support these features as well as automatic parallelism (auto-tasking). The operating system (to be developed within the project) is based on a micro-kernel UNIX standard.
Performance demonstrations are planned at a number of advanced European test sites.
1 blvd Marius Vivier-Merle
F-69443 Lyon Cedex 03
tel: +33 72 35 84 00
fax: +33 72 35 84 10
Forth Research Centre [GR]
Sekas GmbH [D]
Phoenix VLSI Consultants Ltd [UK]
Universidad de Barcelona [E]
University of Edinburgh [UK]
University of Manchester [UK]
SHIPS - 6253, December 1993
please address enquiries to the ESPRIT Information Desk
html version of synopsis by Nick Cook