Re: 1teraflops cell processor possible?

"Nick Maclaren" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
> In article <[EMAIL PROTECTED]>,
> Rupert Pigott <[EMAIL PROTECTED]> wrote:
> >"Robert Myers" <[EMAIL PROTECTED]> wrote in message
> >news:[EMAIL PROTECTED]
> >
> >> No, I don't think it is, either.  What bothers me is what I perceive
> >> to be an enormous misallocation of resources.  Too much money going
> >> into building big machines with yesterday's architecture, not enough
> >> going into more basic research on both hardware and software to find
> >> the best way to address problems in a massively parallel fashion.
> >
> >On the other hand look at IA-64... Huge sums of money in
> >"tomorrows" architecture and yet those clunky old ones
> >seem to stay within touch of it on a tiny fraction of the
> >IA-64 budget.
>
> That is true, but it is only a part of the story.
>
> >The advantage of using stuff you already have to hand
> >when you're working on something new is that it's a known
> >quantity, it's almost certainly cheaper in the short-term
> >(important if you have limited funding), and it allows you
> >to prototype more, erm, accurately...
>
> This is NOT true!  It is very often cheaper to start from scratch,
> and often makes prototyping easier.  It is common for a project

That's a pretty strong assertion Nick. Let's take a look
at the options open to the guys looking for a core for
their spiffy new MPP :
1) Design a new core & bring up a new simulator from
scratch.
2) Utilise a standard core that has been stuffed into
many ASICs that has a bunch of simulators and libraries
already tested and implemented.

Option #2 Definately looks easier for your common or
garden prototyper. This was what INMOS saw when they
embarked upon the RMC before SGS bought them. A friend
of mine whom I lived with for a few months was one of
the guys behind that particular re-usable core.

> that is constrained to use what already exists to spend 75% of its
> resources bypassing problems that simply would not exist if it
> could have started from scratch.

I do accept your point about unnecessary baggage being
a strong risk, but that really should be taken into
account when selecting your core for the job. This is
why there are lots of different shapes and sizes of
core out there.

> The point is that you should always start off VERY, VERY simply,
> because (a) you may get it roughly right, (b) you will have to
> add complications to fix up nasty cases and improve performance
> and (c) the second version will be bigger than the first.  The
> IA-64 project broke this rule, and the transputer one didn't.

The Transputer appears to have been designed by a small
group of people having fun and who did it either because
they cared passionately about it or for the sheer hell
of it.

As a remotely related aside, and hopefully one that won't
lead to any horse-heads in my bed... The in-house CAD system
was called "Fat Freddy" (it's a comic that revolves around
a malignant tom-cat, some hippies and cannabis), it was
written in BCPL (Colin Whitby-Strevens worked for INMOS).

The CAD group did a rather dubious photo of one of them sat
at one of those Fat Freddy machines, stark bollock naked
grinning insanely. No prizes where they got that sick idea
from, I still have nightmares about it. Come to think of
it does anyone know if that infamous photo survived ? :)

Cheers,
Rupert