Author: leonid
Date: 18:13:49 04/22/01
Go up one level in this thread
On April 22, 2001 at 19:32:11, Heiner Marxen wrote: >On April 22, 2001 at 18:19:17, leonid wrote: > >>On April 22, 2001 at 17:36:18, Heiner Marxen wrote: >> >>>On April 22, 2001 at 09:57:10, leonid wrote: >>> >>>>Hi! >>>> >>>>If you want to solve one mate and say your result, then this is the position: >>>> >>>>[D]2R2bNn/Rp1n1k1b/1q1ppB1p/1Np1p1p1/2P1P2P/3P2PB/4Q2K/8 w - - >>>> >>>>If you program says nodes/per/second, please indicate it for this position. >>> >>>According to Chest this is a mate in 7 moves, with 3 solutions: >>> >>>Bxe6+ Kxe6 Qg4+ Kf7 Qxd7+ Kg6 Qf5+ Kf7 Nxh6+ Bxh6 Rxb7+ =*= Nxd6# >>>Qh5+ Bg6 Bxe6+ Kxe6 Qg4+ Bf5 Qxf5+ Kf7 Qh7+ Bg7 Qxg7+ Ke6 Re8# >>>Rxf8+ Kxf8 Ra8+ Nb8 Rxb8+ Kf7 Qh5+ Bg6 Qxh6 Qd8 Nxd6+ Kxg8 Qxh8# >>> >>>On Athlon K7 600 MHz with 350MB hash: 950 seconds, 66,144 nodes/sec. >>> >>>Just for fun I tried this one also with hash table disabled. Timings: >>>depth with without >>># 3 0.02 0.02 >>># 4 0.42 0.42 >>># 5 5.23 6.90 >>># 6 62.19 113.11 >>># 7 950.10 2285.93 >>> >>>Heiner >> >>Hi, Heiner! >> >>It is look like that we have, in general, very close NPS. With Rebel I could see >>that NPS goes somehow down with hash unable. >> >>Salut, >>Leonid. > >Hello Leonid, > >Yes, I would expect that NPS goes down with hash table enabled, since >maintaining the hash costs time, and does not reduce the work per node, >but rather saves complete nodes/subtrees. > >For the above example without hash table my NPS increases to 73848. Hi! In many "ordinary" positions my program will do just like your or slightly higher in NPS. But when the number of pieces are considerable (positions that I like for verification) then my NPS will be well often below number that you said. Somewhere around 40k - 65k. This also was my finding that goes against all "common sense". Very strange!!! Selective search NPS will go usually between 2 and 10 times of brute force search. This I never expected. Some bug?! Could be. Recently I found one in my counting of NPS. Counting was done less, that it should do, just in one ply. There only one position went to count instead of all positions that were used in it. But fault there was not big. Cheers, Leonid. >Cheers, >Heiner
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.