Re: 1teraflops cell processor possible?

On Mon, 1 Dec 2003 08:43:57 -0600, "Del  Cecchi" <[EMAIL PROTECTED]>
wrote:

>
>"Robert Myers" <[EMAIL PROTECTED]> wrote in message
>news:[EMAIL PROTECTED]

<snip>

>> >
>> Never said it wasn't a good box.  I believe I said I thought IBM was
>> proud of the box, and they should be.  Grabbing a bunch of them and
>> lashing them together at LLNL, however, served no purpose whatsoever,
>> though, other than to one-up the ES.
>>
>> Now that you've provoked me, that money would have been MUCH, MUCH,
>> MUCH better spent on basic research to get a handle on how to program
>> these damn things and on something of the kind that DARPA would fund.
>> DARPA would be embarrassed by Blue Gene, too.
>>

>You need to get your timeline straight.  If Blue Gene/L is running today,
>the design was started long before 4/2002.  One doesn't need a crystal ball
>to figure out that all this bio stuff would be a huge market, nor to look at
>the calculations to be made and define systems that folks knew how to build
>and knew how to program that could begin to do those calculations.
>
>Robert, you need to tone down the messianic pronunciations a little.  You
>are not the only smart person looking at these problems.  Just because you
>don't agree doesn't mean the decision was wrong.  And in quite a few cases
>it really isn't a zero sum game.
>
>In my own field I have on several (numerous?) occasions realized, in
>retrospect, that things that were done or not done turned out to be
>reasonable decisions even though at the time I thought they were dumb.  And
>sometimes things I thought were great ideas turned out to  be dumb.
>

My only conern in writing the post to which you are responding was,
"Del is going to take this the wrong way."

If there is a huge commercial market for computers to do biotech
calculations and IBM knows how to make and sell them, then having our
national laboratories buying the biggest and the bestest of what IBM
is selling anyway just doesn't serve any purpose other than making
sure that our national labs have the biggest and the bestest.

Of course I know the project was started before 2002 because I cited a
1999 IBM whitepaper laying out a course of research for such a box.

The meeting in April 2002 was icing: they had to make sure that they
lashed together enough processors to be able to one-up Japan.  Since
the solution is scalable, they could have skipped the travel expense
and meeting time and just gotten out a calculator to figure out how
many of IBM's boxes they had to buy and wire up.

I don't have a beef with IBM.  I have a beef with the DoE.  It's true
that this is comp.arch, not comp.national.policy, but this happens to
be a national policy question of great interest to many of those who
follow comp.arch.

As to it being a zero sum game: it's true, if they hadn't spent the
money on some big high-profile program that would top the Top 500 and
produce lots of color plots, they probably wouldn't have spent the
money at all.

Messianic or not, humble or not, I'm standing by my position: spending
money on monsters like Blue Gene is a waste.  It won't make any real
progress on the problem it purports to be aiming at (I'm repeating
myself), and it would be *much* better to have tossed a few of those
boxes out to universities and let them figure out how to cope with a
few thousand of what Rupert refers to as mice, rather than putting
65000+ of them together to see what comes of it, other than a
meaningless Linpack score.

>So have a little humility, try to be objective.
>
I have the humility to know that when I say something blunt publicly,
I am taking the risk of being proven wrong just as bluntly and just as
publicly.


RM