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

On May 21, 11:01 am, Robert Redelmeier <red...@[EMAIL PROTECTED]
> wrote:
> Robert Myers <rbmyers...@[EMAIL PROTECTED]
> wrote in part:
>
> > That was *exactly* my point.  The proposal was that software
> > development would go to hell in a handbasket if more stringent
> > standards were applied.  The CISC problem has been worked to
> > an extent that no one foresaw, and software verification could
> > similarly be worked to an extent that neither of you foresees.
>
> At what cost and via what mechanisms?  The CISC "problem" was
> not consciously worked as a problem, and certainly not around
> any edicts.  x86 is rather like english -- it succeeded and
> prdominates for subtle reasons _in_spite_of_ theoretical problems.
>
A recent poster to comp.arch referred to the "cubic dollars" that AMD
and Intel had spent to make the instruction decoding problem go away,
so I assume that the investment was large.  The software quality
problem is huge and a huge investment to solve it is more than
justified.

Market mechanisms were adequate to drive a solution to the RISC vs.
CISC problem, but market mechanisms do not always work.  You have only
to look at the pharmaceutical industry.  I urge you not to take me on
about this one.  I lost a good friend three summers ago because of the
long term effects of a miracle drug that was introduced before we had
our current system of regulation.

The system of drug regulation we have now is far from perfect, but a
stroll around Cambidge****t on the edge of the MIT campus will clue you
in that lots of money is being made, in spite of hugely burdensome
regulatory requirements like drug trials.

> Quality is a variable to be optimized, not some
> deity to be wor****pped.  "Beware false gods".
>
I can just imagine hanging that as a motto in someone's office.
You're kidding, right?  Heard of ISO 9000?  Prepare to wor****p a false
god.  From what I've seen of it, I'm not keen on ISO 9000, but the
success of something so clunky should be a clue.

> > Look.  The bald statement was made that financial
> > institutions know how to estimate risk.  Given the moment
> > that the claim is being made, it's beyond ludicrous.
> > It's like claiming that George Bush knows how to run a war.
>
> Merely because something turns out badly does not prove
> any particular cause of that failure.  Ex post facto.
>
It doesn't matter what it proves.  No one will be asking George Bush's
advice about war making except as a learning from mistakes exercise.

Similarly, citing financial institutions as expert in risk management
would have worked before Long Term Capital Management.  As it is, the
financial industry, like George Bush, is a study in the catastrophic
consequences of being overconfident.

The funny thing about risk, like NASA's estimates of risks to the
shuttles, is that you learn that the methodology is faulty only after
a catastrophic failure.  A statement of confidence about handicapping
of risk should tell you to find another handicapper.

> > You can act silly in every way you want.  In every field of
> > commerce *except* software development, it's getting harder and
> > harder to lay risk off onto the end user.  That's the direction
> > that *capitalism* has taken.  These issues are settled in the
> > courts and legislatures, not in Usenet rants.  Sooner or later,
> > laissez-faire software development will be reigned in because
> > the ac***ulated risks to society of the system we have now are
> > unacceptable.  Calling me names will change nothing.
>
> There is risk in everything we do.  And different coping mechanisms.
> I see no reason to suppose that software quality will be legislated
> or otherwise adjudicated.  Legal and scientific proofs are
> fundamentally different.  Far more likely, certifying agencies
> will evolve like UL.

As I've already pointed out, they already have evolved.

> Where human life is at risk, certification
> might be legislated.  Otherwise using the principle of minimum
> interference, contract and tort remedies would apply.  Note UCITA
> failed in nearly all states, and safe harbor in some.
>
Big changes are on the way.  Maybe later rather than sooner, but they
will come.  If you're going to plug into the internet, you will be
regulated.  I'm not happy about that, but I see no way around it.

> > We get it, all right.  People's identities and medical
> > records are stolen en masse, bank accounts are pilfered,
> > and the Internet is home to powerful botnets with unknowable
> > levels of capability or maliciousness of intent.
>
> Oh dear, you really _do_ live in fear.  NYC? My sympathies.
>
> Most of the listed items are _potentials_ whose probability
> and consequences need to be weighed against other threats.
>
There's nothing hypothetical about what I presented.  All the
incidents I've described have already occurred.

Robert.