Author: Albert Silver
Date: 05:57:50 10/10/00
Go up one level in this thread
On October 09, 2000 at 22:31:33, Robert Hyatt wrote:
>On October 09, 2000 at 13:59:55, Aaron Tay wrote:
>
>>On October 09, 2000 at 13:38:37, Robert Hyatt wrote:
>>
>>>On October 09, 2000 at 12:46:27, Joshua Lee wrote:
>>>
>>>>Since you can't compare them by playing each other what about comparing
>>>>printouts? Are there any printouts that have NPS Depth line etc. If i had this i
>>>>would attempt. I did look at a game against Chiptest where CB had 14 Ne6 but
>>>>played Na4<-- not the losing move but not as good as Ne6 CB get's a queen Vs 3
>>>>minor pieces which is not easy but it would be alot better than the situation
>>>>before if CB could keep his queen on and exchange rooks it makes it easier. As
>>>>Black's only chances would be to obtain a new queen not likely. So the outcome
>>>>is probably a draw but this is just my guess. To see this move for Crafty would
>>>>be over 14Ply on my 800Mhz Athlon just to give you some idea. If you can
>>>>elaborate on CB's earlier versions and how they played maybe this will answer my
>>>>questions and help you to improve on crafty.
>>>>
>>>>
>>>>Sometimes it's all about asking the right question.
>>>>
>>>>thank you
>>>
>>>
>>>Actually I did this a couple of years ago and posted the results in r.g.c.c I
>>>believe. I took a couple of the world championship 1986 games (the wc won by
>>>CB for the second time in a row) and had Crafty 'annotate' the games. It was
>>>uncanny how they agreed tactically. Of course, 1986 CB was doing maybe 100-200K
>>>nodes per second max, so crafty had a big edge in speed. But it found _zero_
>>>tactical mistakes by CB.
>>>
>>>I analyzed the games partly in a discussion with Chris Whittington where he
>>>was into the usual mode of criticizing Cray Blitz for any reason. He picked on
>>>one particular move as looking foolish. Someone else pointed out that CSTal
>>>played the _same_ move, as it was tactically forced to avoid losing a pawn,
>>>but of course that didn't make any difference. I became interested in how
>>>Crafty would compare. I gave crafty more time than CB had, using faster
>>>hardware than CB had (1986 hardware for CB, remember), and it couldn't find
>>>any move it would label as a mistake or oversight.
>>>
>>>CB was tactically very strong. In most positional cases the two programs were
>>>in agreement as well, which is not surprising since I wrote both.
>>>
>>>The main advantage CB might have had back then was far faster hardware than
>>>anything I might run Crafty on in 1986. Of course Crafty would have run on a
>>>Cray back then, but it would have been far slower as Crafty is not vectorized
>>>while CB was. I suspect there is not a lot of difference in the two programs
>>>today. CB might have a tactical edge due to singular extensions and a bit of
>>>selectiveness near the leaf positions, while crafty probably has more >positional (particularly endgame) knowledge (excepting king safety where CB >was clearly better).
>>
>>Sorry to interrupt, but does that mean that as of today you wouldn't bet on
>>Crafty running on your current hardware to beat CB (the lastest/last version)?
>>
>>So when do you anticipate will Crafty overtake CB? In 5 years time, when everone
>>is using better hardware?
>
>
>You will have to rephrase your question a bit. But to help, if you meant
>"can crafty on the quad xeon play with Cray Blitz on the T90?" the answer
>is _NO_. CB on the T90 runs at around 7M nodes per second, about 7X faster
>than the quad xeon. I wouldn't want to play such a handicap match.
>
>On the other hand, if you mean "Can Crafty, on the best box you can get today
>play with CB on the T90?" the answer is yes. I have some data from Tim Mann's
>21264a machine which is about as fast as my quad xeon, but using a single cpu
>at 667 mhz. A 16 cpu machine would be faster than Cray Blitz. And I would
>expect it to win more than it would lose, although I think it would be pretty
>close.
On the other hand, CB was originally written for a much older and slower Cray so
two questions come to mind:
1) I presume the difference between the older Cray and the latest and greatest
T90 is more than some overclocking, so mightn't it be _possible_ to write a
better optimized CB, that took advantage of the hardware's evolution, and thus
achieved a much faster node count (among other things) than the current version?
2) You mentioned that even letting Crafty going over CB's 1986 games with
greater time and depth, it was at a loss to find a tactical goof by CB,
suggesting that CB's tactical search was considerably superior. Is this only due
software differences made possible by hardware differences, or do you think it
would be possible to use similar searches/extensions in Crafty today? Would this
hurt it? I know nothing is without a cost, big or small, which is why I ask.
Albert
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.