Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The art of debate

Author: Robert Hyatt

Date: 20:37:35 01/26/00

Go up one level in this thread


On January 26, 2000 at 19:16:39, Tom Kerrigan wrote:

>On January 26, 2000 at 18:51:02, Bruce Moreland wrote:
>
>>Maybe, since it is a chip, it does more than one thing at once.  I do *not* have
>>an EE background, but it would seem to me that you could design a chip that
>>would do several things at the same time, and it would take a very long time to
>>do the same thing serially.
>>
>>bruce
>
>Yup, this is the cool thing about chip design. You can basically do everything
>in parallel, and that's what the DB chip does.
>
>However, when you're talking about a "Deep Blue instruction", I take that to
>mean "telling the DB chip to do something."
>
>Look at it this way: if you had to send the chip 40,000 instructions just to get
>one single node searched, you might as well just use a Pentium.
>
>Bob's claim is so absurd that I could go on for hours listing obvious reasons
>why it isn't true.
>
>-Tom


What on earth are you talking about?  I never said anything about sending DB
40,000 instructions.  I said Hsu might be thinking of the number of different
"things" he does in the chip...  which doesn't necessarily mean that those
translate directly to x86 assembly instructions.  IE when I (using bitmaps)
ask the question "is this pawn passed" I think of that as one operation,
because it _is_ using bitmaps.  In Cray Blitz, it was _not_ one instruction
by a long shot...

You have two different expertises trying to talk with a common language.  I
would not assume that Hsu meant 40K x86 instructions.  I would not assume
he meant 40K gates.  I would not assume he meant 40K "things" on the chip that
get done for each node.  In short, I wouldn't assume anything, I would ask.

I often "guess" but I indicate when I am doing so.  I sometimes know, based on
something Hsu has told me.  I usually also indicate when I am doing that.  But
don't try to put insane words into my mouth and then say "that is obviously
insane as anybody can see" because _I_ didn't say it.

I don't think it is easy to speculate on how many GP instructions it would take
to emulate a hardware design.  First question has to be "which GP architecture"?
I have to think carefully before I would try to estimate how many normal GP
instructions it would take to do some of the things I did in Cray Blitz using
the magic of vector hardware.  IE when I computed mobility, each square had a
different mobility score...  and one instruction was used to extract the right
scores from the vector.  I'd have to think about how to write that in X86
assembly.  And I could easily be off by a factor of 10 if I didn't spend a day
or two thinking about it.

Hence my speculation that 40K was "just a number that sounded sufficiently large
to make the point" and nothing more.  A little thought would suggest that 40K
is an impossible number without knowing the architecture.  40K on a PC might
mean 100K (or more) on a Sparc, for example.  There is too much variability.

40K is _just_ a number.



This page took 0.01 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.