Author: Vincent Diepeveen
Date: 16:24:48 12/23/99
Go up one level in this thread
On December 23, 1999 at 17:18:18, José de Jesús García Ruvalcaba wrote:
>On December 23, 1999 at 16:46:32, Robert Hyatt wrote:
>
>>On December 23, 1999 at 14:59:52, José de Jesús García Ruvalcaba wrote:
>>
>>>On December 23, 1999 at 14:53:16, Robert Hyatt wrote:
>>>
>>>>On December 23, 1999 at 14:43:33, José de Jesús García Ruvalcaba wrote:
>>>>
>>>>>On December 22, 1999 at 18:37:16, Robert Hyatt wrote:
>>>>>
>>>>>>On December 22, 1999 at 17:38:27, José de Jesús García Ruvalcaba wrote:
>>>>>>
>>>>>>>On December 21, 1999 at 17:44:06, Robert Hyatt wrote:
>>>>>>>
>>>>>>>[big snip]
>>>>>>>>
>>>>>>>>Also, chess is _far_ from "embarassingly parallel". It is one of the more
>>>>>>>>difficult-to-program parallel algorithms, because alpha/beta is a strictly
>>>>>>>>defined sequential algorithm. Doing it in parallel invites a lot of extra
>>>>>>>>work that can't be avoided.
>>>>>>>>
>>>>>>>[big snip]
>>>>>>>
>>>>>>> I was just about to begin a new thread asking "is there a quick and dirty way
>>>>>>>of parallelizing a chess search?". By your post I guess that the answer is "no".
>>>>>>>José.
>>>>>>
>>>>>>
>>>>>>There are "quick and dirty" ways to do it. But they don't produce what would be
>>>>>>called stellar performance...
>>>>>>
>>>>>>unfortunately... :)
>>>>>
>>>>> Where can I find about those 'quick and dirty' ways? Poor performance was to be
>>>>>expected, of course.
>>>>>José.
>>>>
>>>>
>>>>The simplest idea is "young brothers wait" which has been around in various
>>>>forms since 1980. It is still non-trivial to handle the locking and potential
>>>>race conditions... but it is not terribily difficult to get up and going.
>>>
>>>Did you use it in 1983 Cray Blitz?
>>
>>
>>Cray Blitz went like this: 1983 (first parallel version, completed in 2
>>weeks when cray surprised us with a working dual cpu XMP) we split at the
>>root only. Typical performance was maybe 1.5x faster in good cases, no faster
>>in some. 1984 saw "PVS" come along (principle variation splitting, not to be
>>confused with todays "PVS" serial search (principle variation search). This
>>was harder, but all processors stayed together at the same node in the tree,
>>so it wasn't too hard. Next (1985) came an enhanced version that eliminated
>>a lot of idle waiting. And finally (1988) I finished DTS which was about as
>>good as can be done, but _very_ complicated. It took me over one year of
>>_full time_ work to debug the thing after I had finished the coding. And
>>I still found a bug here and there every time we played.
>>
>>The current search in Crafty is not as good as DTS, but it is better than
>>the other approaches I used. The main thing I do is that there is very little
>>time where one processor is spent waiting on another for anything...
>
>What does 'DTS' stand for?
Bob i've got proof of bad working of singular extensions.
First deep thought sees because of a horizon effect that c5 move
as i explained some time ago,
then it misses some simple tactics against fritz. The castling
move still is positional determined (though diep plays g3 small depths
already and above and keeps it, so that's nearly blitz level for it to see).
However the c4?? move is a simple tactical blunder, which after some plies
DIEP picks up quickly. DIEP picks it up TACTICALLY. We'll forgive deep
thought being positional worse than any of the programs i wonder here
about how they ever in the world get ranked so high at SSDF.
It should pick up this tactic very easily. I mean let's see how little
plies it is...
moves: 13.O-O? (g3 is much better)
16.c4?? (c4 is a losing tactical shortterm blunder here)
Deep Blue - Fritz WCC95, 1995
1.e4 c5 2.Nf3 Nc6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 e5 6.Ndb5 d6 7.Bg5 a6 8.Na3 b5
9.Bxf6 gxf6 10.Nd5 f5 11.Bd3 Be6 12.Qh5 f4 13.0-0 Rg8 14.Kh1 Rg6 15.Qd1 Rc8
16.c4 Qh4 17.g3 Qh3 18.Qd2 f3 19.Rg1 Rh6 20.Qxh6 Qxh6 21.cxb5 Bxd5
22.exd5 Nb4 23.Bf5 Rc5 24.bxa6 Nxa6
25.Nc2 Qd2 26.Ne1 Rxd5 27.Nxf3 Qxf2 28.Be4 Ra5 29.Rg2 Qe3 30.Re1 Qh6
31.Bc6+ Kd8 32.a3 f5 33.Rc2 Rc5 34.Rxc5 Nxc5 35.Rf1 Be7 36.a4 f4
37.gxf4 Qxf4 38.Rg1 Nxa4 39.b4 Qxb4
0-1
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.