Glenn Cole writes:
>(Don't look for 32-bit computing to go away anytime soon. Mr. Pollack
>has a chart that shows 32-bit chips scheduled to be released in 2003.)
32-bit computing will be around for a long time for the same reason that
8-bit computing is still around: more bits = more expensive, and some
tasks don't benefit from the wider data paths. Fry's now sells Z80s for
about a dollar each, which means that they're probably 50 cents in
quantity. When you can build a whole computer for under $10 at Fry's
prices (most of that is the clock oscillator), you can use them in lots
more places than you can computers where the CPU alone costs hundreds of
dollars.
>To really take advantage of this chip, it appears that compiler technology
>will be all-important. (Remember, the "parallel" part of this architecture
>is "EXPLICITLY parallel." Who do you think has to be explicit?)
Not just the compiler. An explicitly-parallel mainstream computer makes
possible some very interesting application architectures. The Transputer,
a RISC chip from Inmos designed to be used in highly-parallel systems, is
usually programmed with a language called Occam. Occam allows the
programmer to specify data flow rather than control flow, and the
computer(s) execute instructions whenever data from the previous step
becomes available. While that's not what Intel/HP have in mind, exposing
the parallelism at the source level could make for a new and much more
powerful application design methodology.
>An analyst at Dataquest, Martin Reynolds, says:
>
> "The question you have to ask is, 'If this is so great
> why hasn't it been done before?' The reason is that
> the compilers were so scary to do."
>
>I wonder why he uses the past tense, as in "compilers WERE so scary."
Because we've had 15 years of experience with optimizers that do
instruction scheduling. The algorithms are well-understood,
well-characterized, and simple to implement -- at least to today's
programmers.
Science and technology move exponentially because each generation of
scientists and engineers gets to start with the knowledge and technology
provided by the previous generation. The tough engineering problem of 10
years ago is just a question on a computer science midterm today. A 1980
landmark Ph.D. thesis becomes 1997 freeware.
It's interesting to go back through the history of computer science and
see how these tools developed. A real eye-opener in this respect is
Andrew Hodges's biography of Alan Turing, "Alan Turing: The Enigma."
Hodges uses Turing's notes to trace the development of several
programming concepts. The idea of a subroutine seems intuitively, even
painfully obvious. But Turing worked on this concept for months, until he
figured out how and why to do it. Somehow, it wasn't intuitively obvious
back before anyone had ever thought about it.
A paper in this week's _Science_ (<http://www.sciencemag.org>) describes
a milestone in computing and chemistry: researchers specified the shape
of a never-before-seen protein molecule (the shape is what makes proteins
work), fed the shape to a computer program they'd developed, and after
many hours of processing on a large network of superminis, the program
produced the sequence of amino acids necessary to create a protein in the
specified shape. When they synthesized the protein according to the
computer's instructions, it had exactly the desired shape.
This has been a goal for 30+ years, and it's finally been realized. Does
anyone doubt that a similar program will be available as freeware in 10
years, and that it'll work in a few seconds on a computer you could buy
at Sears? And that the concepts and algorithms used in the program, the
fruits of 30 years of research, will be second-nature to the next
generation of software engineers? Twenty years from now, a biochemist
who's just starting first grade today will say to her colleague over
lunch, "I can't imagine what it must have been like in the old days, when
they just had to search for proteins by luck."
-- Bruce
--------------------------------------------------------------------------
Bruce Toback Tel: (602) 996-8601| My candle burns at both ends;
OPT, Inc. (800) 858-4507| It will not last the night;
11801 N. Tatum Blvd. Ste. 142 | But ah, my foes, and oh, my friends -
Phoenix AZ 85028 | It gives a lovely light.
btoback AT optc.com | -- Edna St. Vincent Millay
Unsolicited mail to [log in to unmask] will be inspected for a
fee of US$250. Mailing to said address constitutes agreement to
pay, including collection costs.
|
