Re: 1teraflops cell processor possible?

"Robert Myers" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
> >
>
> And why would anyone fool with risky streaming architectures when you
> can build the world's biggest computer with Power4's running at
> 750MHz?
>
> Obviously, you wouldn't, if you can get the powers that be to buy into
> and/or sell the idea that said computer will be able to do the
> impossible.
>
> Sorry, Del, but Blue Gene makes about as much sense as the Space
> Shuttle did.
>
> RM

Actually the space shuttle made sense.  It just didn't work out like
they envisioned.  It was to be a prototype for a next generation of
reusable orbital vehicles, as I recall.  Instead folks lost interest and
the prototype became the final.  How like computer architecture is that?

As for blue gene, and recalling that I am a circuit designer and not an
architect, it is an initial member of what might be a family of
massively parallel supers.  Why are you so convinced that some bizarre
streaming architecture is the answer, and massively parallel systems are
dinosaurs or kluges?  At least that is what I interpret your reference
to the shuttle as meaning.

If the idea is to compute protein folding, then maybe using like a
million processors is in fact the way to go.  And 128K processors is a
good start.  Cray is gluing opterons together.  IBM glued 690's
together, now they are gluing way more 440's or whatever.

I guess the proof of the pudding will be in the eating.  We'll see if
useful science gets done.  And the alternative proposals will get funded
if they can convince folks that they can do useful science.  IBM would
even help them, subject to satisfactory terms and conditions.

del