Re: Computing Editorial (please read)

[Michael Joosten <joost@c-lab.de>]

Lee Hart wrote:
> 

> 
> For example, analog computing has languished for decades. They could
> always do simulations and solve complex differential equations far
> faster than the best digital computers. Built with modern chips, analog
> computers could put us an order of magnitude ahead for certain types of
> problems.
> 

Weighing noise effects, temperature coefficients and aging against
rounding errors in fixed and also floating point signal processors? 
Well, finally you need to feed the results into digital from, so...

> Another: We have chosen the "god in a CPU" architecture for our PCs --
> one supreme all-powerful all-knowing CPU that runs everything,
> surrounded by much smaller CPUs that are total idiots in comparison.
> Instead, what about architectures where there are large numbers of small
> CPU, roughly equal in capability, networked together to form a powerful
> system.

Because, when Intel was dominating the PC processor industry, and few,
if any, stake in the DSP and special processor market, they avoided and
precluded together with Microsoft any 'deviation' of the divine religion
of 'One big fat CPU does it all'. Remember the beginning of the
multimedia PCs? In the beginning, there were many attempts and even
products with dedicated video processors, MPEG decoders, sound
processors of which only sound survived (and even that was later
attacked with some half hard/half soft approach in some chipset for)
cheap motherboards. 

Now, they (the guys in the research depts. at least) have realized that
the power density of today's chips is rapidly leaving the hot
boilerplate denisty and nearing the that of the outer sun. No wonder
that they step back to think again about multiple processors, but as
hyperthreading has already shown, there still needs much to be done to
make applications correctly aware of this...


 Working examples of such systems can outperform supercomputers,
> at far lower cost. They could readily be built with Z80-type SBCs by
> hobbyiests.
> 

But mostly only in special situations. Fast(latency!!)/wide
interconnects are still the major problem. Let's see how InfiniBand and
Hyperchannel etc pp. will fare.
Oops, sorry, wrong processor class...

Banding small processors together sound more like the original
Connection Machine (though CM-1 and 2 used only single bit processors,
and CM-5 jumped to SPARC nodes...) or Transputers. The major problem
here is that most 'mundane' jobs are actually quite serial in nature, so
it is often hard to actually parallelize (?) them efficiently.

> There is a tremendous amount of new territory to explore. And, it is
> more likely to be explored by hobbyists and experimenters, not by large
> corporations who have largely dismantled their R&D efforts in favor of
> profits this quarter.
> 

Hmm. Messing around with hundreds of Transputers with just 4MB per node
was no real fun eben in it's heyday. Well, it was fun, I guess, but if
you had a specific problem, allocating them correctly, and feeding them
the work were the most difficult part. 


> 

> 
> > mostly these inventors were *not* part of the main stream... and
> > they were *not* plugged in to the "that's the way we've always
> > done it" mentality. That's why the airplane had to be invented by
> > bicycle mechanics.
> 
> Absolutely!!! Who do you think will invent the next generation of
> computers. Microsoft? Intel? Not a chance! They will perpetuate the
> status quo.

In some cases yes, in others not. If have to cover a broad spectrum of
topics to get a new idea into reality, plus lots's of regulations
around, you are doomed to use more-or-less large groups. If it is
targeted approach, a few geniusses will suffice. Say Seymour Cray and
IBM's Watson's comments about what he saw in that CDC lab? I'd guess he
would have had used far more designers if he had to care with lots of
specific transaction processing and I/O devices, as IBM had to do.


> > In kit form, this might be a sellable item. The same could
> > be said for building 8-bit microcomputers and microcontrollers.
> 
> But as far as I know, there are no kid's "Electronic Labs" kits with a
> microcomputer. Hey, maybe that's the place to start!

I vaguely remember seeing some of them in .... mid 80's ? Some small
4bit microprocessor box (prebuilt, of course), similar to the older
KIM-1 and such designs (numeric/command keys, some LEDs, some seven
segemt displays).

-- 
Michael Joosten, SBS C-LAB, joost@c-lab.de
Fuerstenallee 11, 33094 Paderborn, Germany
Phone: +49 5251 606127, Fax: +49 5251 606065
C-LAB is a cooperation of University Paderborn & SIEMENS