Re: AMD planning 45nm 12-Core 'Istanbul' Processor ?

On May 12, 11:31 am, Sebastian Kaliszewski
<s.usun...@[EMAIL PROTECTED]
> wrote:
> Robert Myers wrote:

>
> Who cares what some clueless newsgroup poster thinks.
>
In my years on this planet, and, more recently in dealing with usenet
hotheads, I've learned not to rely on my own knowledge and experience
when I don't have to.

Even so:

1. I am not clueless.
2. The people I talk to are even less clueless than I am.


> 1. very expensive
> 2. does not guarantee total correctenss -- it only reduces chances of
error

Given the choice between something with solid mathematical foundations
and something that relies exclusively on someone like you, I'd choose
the former every single time.  As to the cost, when I said we'd be
better off with less software, I meant it, and I still mean it.

You don't have to rely on my "clueless" opinion.  I've posted on the
subject of the costs of our current setup, with citations.  I've had
public discussions with software development managers in critical
applications who lamented exactly as I did.  Well he should have, a
software boner had just cost his high-profile publicly-traded company
big bucks.  As it is, with an internet, *everything* is a critical
application.

Know Edsger Dikstra?  Want to call him an idiot, too?  Just to be
sure, I googled on your name to see if you, too, had won a Turing
Award.  If so, google missed it.

>
> No, I'm relying on cost of software production vs cost's of errors times
> times their expected probability of occurence during lifetime of the
> software.
> Making software immune to errors costing 1e+9$ with probability greater
that
> 1-1e-4 (that is four nines) during 10 years lifetime should not cost
more
> than 1e+5$ (that's about 1 man-year spent on improving quality). If it
> costs more it's time to cut corners and increase error probability
reducing
> production cost.
> And even then it's pointless to run that software on hardware with fault
> probability worse that that of software (i.e. not for home use, nor
small
> company -- your standard PC will die with probability well above 0.5 not
> below 0.0001 within the timeframe (even if you're doing regular
> professional system maitenence and exchange all parts on planed
shedule).
> Then all of that must be immune to operator errors -- also impossible in
> home or small and middle sized business environment.
>
You don't need to recite to me justification for the grubby economics
of software development.  Everyone lives with it.  The part of the
calculation you don't want to do is the risk that's pushed off onto
someone else in the indefinite future.  If you were forced to buy
insurance against that risk, you'd damn well do what I've proposed.
As it is, you can afford to let it slide, because the risk-taker is
always the end user.  That's how this shameful business is set up, and
I wish you'd quit bragging about how good you are at it.  There are
good loan sharks, too.

Robert.