Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty 19.20 source

Author: Robert Hyatt

Date: 20:45:54 08/25/05

Go up one level in this thread


On August 25, 2005 at 23:42:33, Robert Hyatt wrote:

>On August 25, 2005 at 15:43:30, Vincent Lejeune wrote:
>
>>On August 25, 2005 at 11:33:19, Robert Hyatt wrote:
>>
>>>
>>>
>>>On August 25, 2005 at 06:14:11, Bernhard Bauer wrote:
>>...
>>>>I am not sure either.
>>>>If I would make only one blunder in 20% of my games, I would not think it's
>>>>often. However, if I would blunder 20% of the moves in one game, that would be
>>>>quite often.
>>>>Look at the following 3 positions:
>>>>8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w       Fruit2.1 needs 2:20 min to find Kb1
>>>>7k/2p5/2p5/2P4p/3P1p1p/5P1P/8/K7 w    Fruit2.1 gives Kb2 whithin 5 min
>>>>2k5/2p2p2/2p1p3/2P1PpP1/1p1P1p2/1P3P2/1K3P2/8 w
>>>>Fruit2.1 needs 55 sec to find Kb1. My crafty version only 15 sec.
>>>>
>>>>Let me recall something Bruce Moreland wrote: "Search should find it"
>>>>Agressive pruning makes your engine faster and usually stronger, but not more
>>>>reliable.
>>>>Kind regards
>>>>Bernhard
>>>
>>>
>>>If you are interested in test sets, I can run most anything.  I am currently
>>>running the Nolot positions for 60 minutes each.  On the opteron box, a 1-cpu
>>>run gets five correct.  Will see if the 2-4-8 cpu runs find more...
>>
>>Cool , we will be able to compare with results of DB2 (about right and wrong
>>solutions)
>>http://chessprogramming.org/cccsearch/ccc.php?art_id=150640
>
>
>I'll put all the logs on my ftp box when they are done.  Right now, for the
>first 8 cpu test run, here's what's happened to this point:
>
>position 1  Nxh6 is found in 1:06...
>position 2  Rxc5 is found in 8:31...
>position 3  Nxg5 not found in 20 minutes...
>position 4  Nxe6 found in 11:35
>position 5  e5 not found in 20 minutes...
>position 6  axb5 not found in 20 minutes...
>Position 7  Rxd8 found in 12:45
>position 8  Bxh7 not found in 20:00
>
>from this point forward, I only have 4 processor results, the first 8cpu run has
>not yet completed.  I'll give the 4 cpu results from here on and then update
>this later.
>
>position 9  Ng5 not found in 20:00 (it chooses Nxh6 here)
>position 10 Rxf7 found in 1:02
>positoin 11 Rxh6 found in 0:17 (17 seconds)
>
>Total (so far, 6 correct, 5 wrong.  Doubt this will change after the 8cpu test
>completes, but we'll see...


I should add that I lied about the time.  :)  I ran the one-cpu test for 60
minutes per position, but I ran the 4/8 cpu tests for 20 minutes each as that is
enough to get as deep as the 1 cpu test.  I was only running this for the SMP
speedup data, not to see how it would do in terms of getting more correct with
additional search depth due to additional search power.  So all of the above
results were for 20 minutes each, max, not 60 as I originally said.  I had
forgotten about the reduced time limit with more cpus to try to keep the overall
run time to something reasonable...  Since I run each 4 cpu test 8 times so that
I can get a good average speedup number...



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.