python a bust?

[Patrick Maupin]

"Brandon J. Van Every" wrote:
> I have seen the DEC Alpha CPU torn out from under me.  As far as I'm
> concerned, any hardware / language / API / OS has enemies, and those that do
> not market themselves properly are endangered.  Considering that Python was
> available before Java, it is not the success story that it could or should
> be.  The ugly truth of high is it's 1/3 technology and 2/3 marketing.  If
> you believe otherwise, then you haven't had Intel or Microsoft hand you your
> ass yet.

I don't dispute that Python could/should do better against Java.
(I'm actually pretty agnostic on this statement.)

However, comparing hardware and software for cost/benefit is much
worse than comparing apples and oranges.  There are tangible, huge
costs associated with fabbing and selling a chip.  If you can only
sell a few a year it's simply not worth it.  Especially if you
chip requires additional support chips which are no longer sold
because it's not worth it for them, either.

For software, open source makes the economics even sweeter.  You
can often easily justify the cost of paying to incrementally
improve a package you use based solely on your own needs.  For
chips the economics are _way_ different.  Assume for a moment that
the Alpha was open-sourced, and you wanted to create a "modern"
version of it.  Are you willing to spend a half-millon dollars
on tools, and another half-million or more on a mask set to be
able to produce a 90nm version of it which won't even work with
any of the existing support chips because the IO cells on your
fancy new chip aren't even 3V-tolerant?

Bottom line:  the probability of long-term availability of and
support for Alphas tends toward 0, while the probability of long-term
availability of and support for Python tends toward 1  :)

Pat

P.S.  The hardware economics _are_ currently undergoing a radical
change.  If your requirements do not include cutting edge speed,
you _can_ build onesies/twosies using FPGAs for hundreds of dollars,
or even in some cases tens or hundreds of units for tens of dollars.

Or if you really only want a few fast ones, you could forego the
cost of the mask set, and "only" spend a half-million or so for tools,
and fifty to a hundred thousand for a few die on a "multi-project
wafer".

The future may hold "direct write" systems which do not require any
mask set.  If these become practical, the cost of the software tools
will become a much bigger proportion of the total bill, but the number
of potential projects will skyrocket, so history and simple economics
show that competition will cause the tool prices to drop like a rock.

Once that happens, you may very well be able to build your Alpha
chips on demand :)