[HARDWARE] Massive computing power

Raetsel raetsel at home.com
Thu Sep 24 00:19:13 EDT 1998


Okay folks... I'm new here, so maybe this has gone the rounds -- but here
goes:

    We're all working on a massively math-intense project here, regarding
this there can be no argument.  We have checked over a half-million trillion
keys, and yet exhausted only 2.5% of the possibilities.  I'm not saying
anything new here, you all know this from the stats pages.

    What I'm questioning is our individual approach.  I am running a dual
PII/375 (333s pushed to a 75MHz bus clock) under NT 4.  I regularly get 2.08
Mkeys/second.  Okay.  Fine.  Dandy, even.  If we're to come close to the 14
hour challenge that looks to be the next hurdle, there needs to be
innovation on a scale other than what the commercial sector can (or is
willing to) provide.

    I'm not a theorist.  I'm not a computer scientist -- though I do play
one at work.  I don't pretend to understand the math that goes into making
encryption work.  Hell, I don't even know the math that makes it work.  But
(obviously) this has to be a repetitive computational task that is
performed; the only thing that changes is the key we're trying in the lock.
And that kind of work can be hard-wired.

    As evidence, I present an article from Discover magazine, June 1997,
titled The Star Machine.  (go to http://www.discover.com/archive/index.html
Under Search for Keyword in Article enter "gravity pipeline")

    Granted, they put $2 million into this thing.  However, does anyone
think we can come up with a ~$500 PCI device that'll do greater than, say,
25 MKeys/sec?  Something like a crypto-coprocessor?  It'd also have to play
well with others (work in multiple board arrays).  That would be the
equivalent of over 10 dual PIIs (or Celerons).  That's the way I think this
contest would be best approached.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.distributed.net/pipermail/hardware/attachments/19980923/976020ed/attachment.htm


More information about the Hardware mailing list