🚀 go-pugleaf

In article <49600@ames.arc.nasa.gov> lamaster@athena.arc.nasa.gov (Hugh LaMaster) writes:
>In article <6182@amelia.nas.nasa.gov>, eugene@wilbur.nas.nasa.gov
>(Eugene N. Miya) writes: 
>
>|> The solution in marketing is "price-performance."
>
>
>Price/performance is now an issue, now that smaller is cheaper, at least when
>you account for it one way.  It is even more nebulous, though.  And, I
>don't think
>the average man on the street thinks of a collection of Ford Escorts as
>a sportscar because they are "cost effective".
>
>winning in the Linpack race.  That could happen Real Soon Now, but it hasn't...
>
Hmmmm... what happens if I take a 64K CM-2 with floating point and 4 Suns 
frontending it.. sun4/490s. Connect these suns with a reasonable FDDI 
network and populate them with lots of memory 196 MB max. The CM-2 should
have large memory (4GB). I would think this would win the price performance
race and even do well on the Linpacks. The Sun(s) would now be the 
Supercomputer.. given that the CM is really a backend array processor and
the Suns have all of the user-lever software etc. I guess it is a hybrid.

But not necessarily all SIMD.. even at the CM level with slice-wise processing
and the sprint nodes...

And what about 'The network is the system' ...???

-- Peter


-- 
Peter Brewer             ||||     |||||  |||||||||  ||||||  //|||||\  ||||||
lerici@super.org	 ||       ||__   ||     ||    ||   ||           ||
THE Supercomputing       ||       ||     ||^^^^^^\\   ||   ||           ||
Research Center ~~~      |||||||| |||||  ||       || |||||  \\|||||/  ||||||