Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!uwm.edu!ux1.cso.uiuc.edu!ux1.cso.uiuc.edu!m.cs.uiuc.edu!kale
From: kale@m.cs.uiuc.edu
Newsgroups: comp.arch
Subject: Re: parallel systems vs. uni-proces
Message-ID: <3300081@m.cs.uiuc.edu>
Date: 26 Oct 89 16:31:11 GMT
References: <89@<RANG>
Lines: 23
Nf-ID: #R:<RANG:89:m.cs.uiuc.edu:3300081:000:899
Nf-From: m.cs.uiuc.edu!kale    Oct 25 23:05:00 1989


According to NCUBE glossies in front of me: 

"512 billion bytes of memory capacity"
	So up to 64 Mbytes/processor..

about price: 
"... at less than $1000/MFLOP and $1000/Mbyte"

and elsewhere: ".. varying number of processors - from 16 to 8192 -
	that deliver from 8 to 27,000 scalar MFLOPS"

(So, price can be deduced.. But I believe it will be highly non-linear..)

On another point: Yes it is more difficult (but not too difficult)
to program distributed memory machines. But the rewards are great, 
and the challenge is stimulating. I believe they will be
useful for a large fraction of applications.
When people claim that "they are good for some applications, but not most",
they sometimes may mean a particular algorithm or a particular way of doing 
things may be difficult on message passing machines.
But not the end application - it can often be reformulated so it works fine
on them.