[nSLUG] Re: 1, 2 or many? -- how many cores will you want...

Mike Spencer mspencer at tallships.ca
Fri Mar 23 23:22:18 ADT 2007


> A number of CS profs have been commenting on the difficulties of
> writing software for highly parallel hardware.

At least in part and despite the likes of Hillis, Feynman & Wolfram on
the team, that was what caused the Connection Machine to founder, was
it not?  They found that parallelism could do unexpected and stunning
stuff but they practically had to re-invent the architecture for each
new problem or application.

OTOH, Thinking Machines preceded the resurgence of interest in neural
nets and that's a place where parallelism is just the natural way to
go.  Machine pattiducking. [1] (Altho I just learned from google that
Hopfield did a presentation for TM and the TM folks did an
implemntation of Hopfield nets. See

   http://www.longnow.org/views/essays/articles/ArtFeynman.php.)


- Mike

[1] For those who've read Stand on Zanzibar :-)

-- 
Michael Spencer                  Nova Scotia, Canada       .~. 
                                                           /V\ 
mspencer at tallships.ca                                     /( )\
http://home.tallships.ca/mspencer/                        ^^-^^

!DSPAM:46048bc748031665021011!




More information about the nSLUG mailing list