dana@rucs.faculty.cs.runet.edu (Dana Eckart) (04/05/91)
I remember reading in a journal article somewhere that Toffoli and Margolus
(I think) were working on a CAM-7 machine (successor to the CAM-6). If I
remember correctly, the CAM-7 was to have 1000 x 1000 x 1000 16 bit cells!
Has anyone else heard of this (or have I been dreaming again)? If so, can
someone give me some pointers on where to look to find out more about it?
Thanks in advance.
--------------------------------------------------------------
J Dana Eckart INTERNET: dana@rucs-gw.runet.edu
SNAIL: 15 Grandview Dr/Radford, VA 24141biafore@theory.lcs.mit.edu (L. Michael Biafore) (04/07/91)
In article <1991Apr5.134658.6196@rucs.faculty.cs.runet.edu> dana@rucs.faculty.cs.runet.edu (Dana Eckart) writes: >I remember reading in a journal article somewhere that Toffoli and Margolus >(I think) were working on a CAM-7 machine (successor to the CAM-6). If I >remember correctly, the CAM-7 was to have 1000 x 1000 x 1000 16 bit cells! > >Has anyone else heard of this (or have I been dreaming again)? If so, can >someone give me some pointers on where to look to find out more about it? > >Thanks in advance. > >-------------------------------------------------------------- >J Dana Eckart INTERNET: dana@rucs-gw.runet.edu > SNAIL: 15 Grandview Dr/Radford, VA 24141 The article you are referring to may be "Cellular Automata Machines: A new environment for modeling", by Norm Margolus, in Proc. of the 1988 Rochester Forth Conf., or "Cellular Automata Machines", by Norm Margolus and Tom Toffoli, in "Lattice Gas Methods for Partial Differential Equations", Proc. of the Santa Fe Institute, edited by Gary Doolen (1987). However, CAM-7, first described in Norm's thesis, was not built. It has evolved into CAM-8, which will have 4M 16 bit sites per module and as many modules as you like. The sites can be grouped in 2, 3, or more dimensions until you run out of sites, but the length of each dimension must be a power of 2. The entire cellular space can be updated about five times per second. If a smaller segment of the space is scanned, say a 512x512 region, video speeds can be acheived on a dedicated external monitor. If the space is displayed on the host, which will be a Sparc station, display will be somehwat slower. We expect to send the design off for fab this summer, and a basic single-module card SHOULD be available around the first half of 92. A contract manufacturer has not been selected yet (it won't be Systems Concepts), so one could only guess at price. I'd guess 3-10k per module. -- Michael Biafore | ARPA: biafore@im.lcs.mit.edu MIT LCS, Room NE43-354 | 545 Technology Square | (617) 253-5884 (work) Cambridge, MA 02139 |