Author: Vincent Diepeveen
Date: 12:33:01 07/28/03
Go up one level in this thread
On July 27, 2003 at 14:28:09, Tom Kerrigan wrote:
>On July 27, 2003 at 07:41:02, Vincent Diepeveen wrote:
>
>>On July 26, 2003 at 18:17:40, Tom Kerrigan wrote:
>>
>>>On July 26, 2003 at 17:22:02, Russell Reagan wrote:
>>>
>>>>On July 26, 2003 at 16:25:37, O. Veli wrote:
>>>>
>>>>>Since it is hardware, can
>>>>>we expect to be stronger than top software?
>>>>
>>>>I would expect it to be slower than top software, because cpu improvements
>>>>happen so quickly, and FPGA programming (from what I've heard) is not a simple
>>>>task. If he spends another two years working on it before releasing it (as
>>>>Slater said), just imagine how much faster the cpus will be by then.
>>>>
>>>>If you're talking about something massively parallel like Deep Blue, that is one
>>>>thing, but a single PCI card? I doubt that is going to do any better than break
>>>>even with top of the line hardware, so why bother? IBM threw so much hardware at
>>>>the problem that desktop cpu improvements wouldn't catch up for a LONG time, but
>>>>a single PCI card doesn't seem to be worth the trouble of programming the thing,
>>>>because desktop/server cpus will probably outperform it before too long.
>>>
>>>You're comparing apples and oranges here. You're deriding something because it's
>>>a "single PCI card" but you could put a DB chip on a single PCI card, run it at
>>>~25MHz, and it would search 2M NPS and possibly do quite well vs. commercial
>>>programs. And this is an ancient chip. FHH says that shrunk to 0.18um (not even
>>>90nm or 130nm) a single chip on a single PCI card would be faster than Deep Blue
>>>1.
>>>
>>>-Tom
>>
>>Grow up tom. Some gnuchess evaluation from 1997 is going to get butchered
>>massively.
>
>And I'm sure you know _exactly_ how DB's eval function worked, even though it
>hasn't been made public anywhere.
a) We have an exact description in the article of artificial intelligence.
The exact implementation of 1 pattern we do not know, but the number of patterns
is dissappointing.
b) we have 6 games against kasparov. From those the first 5 games reveal loads
of data.
I will list some real horrible positional moves here from which nowadays
programs hopefully will not make much from. From my 'badmoves.txt' i just quote
game 1, otherwise posting gets just too long here.
Bad moves played by deep blue according to Fide Grandmaster Seirawan. Journal of
ICCA june 1997. comments
by Fide Master Diepeveen:
Game 1 :
3 .. Nd7? this is a bad move. Better is Bxf3 followed by e6. We cannot blame
Deep Blue for this mistake. The opening is simply very hard to
play.
Having the bishop pair versus a doubled pawn is not easy. Let's be
clear
here that it is amazing to see it initially wants e6 even which is
a *real* bad move.
Search depth deep blue had is 10 ply here.
10 .. h6? This move is really very bad and i could not explain it very well
until i saw
Seirawans explanation and also Gnuchess and Zarkov play this. All
for the same
reason. Deep Blue loves bishops. Nothing wrong with that. But
weakening king
safety? This move when played at high GM level is enough to get a
bad position now...
The most amazing thing is that black has the chance now to put a
pawn in the center.
How can Deep Blue miss such an elementary move? Playing h6 is very
bad for the
rest of the game but failing to play e5! here is real real bad!
Search depth of deep blue here is 11 ply.
11 .. Qa5? This move is really very bad. The queen has nothing to do there.
Seirawan again guesses
correctly here. Gnuchess 4.0 at the time just counted squares queen
attacked. At a5 it
attacks more. I don't need to mention that Zarkov 4.xx (follow up
of gnuchess of course private
ware thanks to John Stanback for getting a copy) plays also this
move.
Search depth from deep blue here is 10 ply. It wasn't finished at
all.
12 .. Bc7? This move is real bizarre. Not only it loses a tempo it shuts off
the way for the
queen. I don't need to mention Zarkov at the time also played this
move at 10 ply.
Search depth 10 ply. Seems finished to me.
13 .. g5? This move is LOSING. So where the position is already very bad for
black and playing h6
earlier in the game means that you will play g5 here too usually,
only a few top engines
in 2003 do not play g5?? here. DIEP does not even consider it. Some
programs known for a weak
king safety though do play it. In general it requires a big depth
to see that g5 is bad.
Deep Blue finished 10 ply here. Busy searching PV from 11.
22 .. g4? Where kasparov had not done much effort so far to win the game. He
just 'nullmoved' all game;
Later he would say: "I should never have won that first game so
easy, then i would have taken
this match serious".
After g4 here black can resign. Still Deep Blue with the usual
computer effort manages to
make a tactical fight by opening the position. Kasparov should have
been warned here about its
aggressive way of opening positions...
33 .. Qb5? Seirawan considers this a very bad move. I do not care however
because the position is dead lost.
The motives for finding this a bad move is a human one which
objectively is not relevant IMHO.
That's already 7 bad moves in 1 game. How can you still take this program
serious? You sure you want to see more
horrors?
I made today in my game 3 bad moves and i was disgusted because of that. I'm
2310 FIDE rated (don't compare that to
USCF).
>>According to Donninger as explained publicly (but not in english) his move
>>generator is like a part only of the gates that DBII used. In his opinion the
>>design of DB chip was a joke from many perspectives.
>
>Well it doesn't exactly matter, does it? The DB chips _worked_, didn't they?
>
>>Things like searching 4 ply in hardware is *not* going to work simply because
>>the performance loss is too big.
>
>Because...? What's wrong with 4 ply searches in hardware?
>
>>Deep blue ordered moves near to random compared to modern software. Not even
>>killermoves.
>
>I would hardly call MVV/LVA random. You can have ideal move ordering and it
>won't do _THAT_ much better because MVV/LVA is often ideal.
>
>>Then they forward pruned in hardware.
>>
>>Forward pruning in hardware based upon 1 or 2 line pruning rules?????????
>>
>>Do i need to go on?
>
>No, you've made enough of an ass of yourself already. My point was that a single
>chip on a single PCI card is capable of hundreds of millions of NPS, not that DB
>is awesome. If you want to waste your time arguing that DB was crap because of
>your own uninformed opinions, do it somewhere else.
>
>-Tom
This page took 0 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.