Re: [RSA] Multi-Processor Pentium Clusters (was Re: First place.)

Jonathan M. Bresler (
Thu, 17 Apr 1997 15:50:55 -0700 (PDT)

Stuart Stock wrote:
> On Thu, 17 Apr 1997, Warner Losh wrote:
> > I have been hearing for years about multiple cpu designs going on at
> > various government labs. I recall that Sandia had an experimental
> > Pentium MP machine with either 1,000 or 10,000 processors (my memory
> > is failing me) in it with private and shared memory for each (it is
> > hard to tell for sure, cine it was a popular press report). This key
> > thing is just the ticket for something like this. The picture they
> > had in the paper showed a "rack" sized box (that is 19"wide, by 8'
> > tall by about 30-40" deep) full of backplanes and such. I don't know
> > how much that press report got right, but it does make you think.
> One of the efforts you are refering to is the Beowulf Project. A
> Beowulf is a collection of semi-autonomous Linux based PC nodes connected
> via multiple switched and/or channel-bonded Fast Ethernet connections.

perhaps, perhaps not. i imagine that Warner was remembering
dimly a url i posted to the FreeBSD mailing lists about a
1.8 TeraFlop machine. a couple gigaflops arent interesting
to the compuational fluid dynamics crowd. (remember the press
reports about simulating nukes instead of exploding them.)
take a look at
what is your PGP key length? how cheaop are P6's?
the development contract of this computer was "only"
$46 million. aircraft carriers are more than a billion.
B-2 bombers are near 500 million (? memory glitch).

the first model is always the most expensive, later copies will be
/are significantly cheaper.