Author: Uri Blass
Date: 06:46:30 01/13/02
Go up one level in this thread
On January 13, 2002 at 08:41:57, Gian-Carlo Pascutto wrote: >On January 13, 2002 at 08:29:35, Uri Blass wrote: > >>On January 13, 2002 at 08:07:49, Gian-Carlo Pascutto wrote: >> >>>On January 13, 2002 at 07:05:16, Tina Long wrote: >>> >>>>Gigantic hash tables for very quick time limit didn't suit Gandalf? >>> >>>I don't think so. The Athlon can clear 200M in a split second, and >>>it plays 1 0 games fine with this setting. >>> >>>>Gandalf is more knowledge based so GambitTigger looked a lot deeper because of >>>>the quick time limit? >>> >>>Mm. Perhaps. Gandalf doesn't search very deeply, Tiger does. This won't >>>change when playing slower games. Gandalf will search a ply deeper, but >>>so will Tiger (and perhaps two). >>> >>>>The extra Tablebases helped GT? >>> >>>There's no practical advantage to using 6 man >>>tablebases. (according to Robert) >>> >>>>Sample size too small? >>> >>>24 games isn't much, but it's still a huge score difference, >>>enough to be significant. I'll plug the result into elostat >>>and see what comes out. >>> >>>>Did you notice any explaining factors or do you think this is a fair comparison >>>>of strength? >>> >>>I think it's fair comparison, the hardware was about equal and both were playing >>>at full strength. If there is something wrong with my setup >>>(I was operating Gandalf), then I honestly wouldn't know what is was. >>> >>>>Apriori I would have expected about 8-6-10 or thereabouts. >>> >>>Hmm. I'm disappointed with Gandalf so far. I used to run Fritz 7 on the same >>>account and with Gandalf the rating has dropped about 70-100 points! >>> >>>It doesn't seem to be a real top program to me. I wonder what the SSDF result >>>will be. >>> >>>-- >>>GCP >> >>I will try another possible theory(I do not have gandalf5 so I cannot check it) >> >>Maybe gandalf need time to get used to big hash tables for some reason. >> >>The fact that it does not lose on time even on 1 0 games does not contradict >>this theory because it is possible that gandalf starts by searching 50 knodes >>per second in the first seconds and only after a minute the number of nodes per >>second becomes 300 knodes per second and the time that gandalf needs to get used >>to big hash tables is bigger than the time that it needs to >>get used to small hash tables. > >I have no idea why or how you think this is possible. It would make the >chessprogram practically useless! > >It's wrong. I see it is by looking at the statistics window. Gandalf does >about 300-400 (800 in some endgames) knps on my system, and that number is >fairly constant. > >Moreover, also by observing the stats window, I do not have the impression >that Gandalf clears it hashtables frequently, so I do not think they had >anything to do with it. > >-- >GCP I thought that it may be possible because I know that the nodes per second for gandalf4.32h as a chessbase engine was not constant. I also know that it is possible even with Deep Fritz. It never happens when I try to use 48 mbytes but for some reason it happens sometimes when I try to use 64 mbytes. I do not know if it is because of other applications that the computer use at the same time or because of windows2000 but it never happens in another computer (p850 with windows98). The case with Deep Fritz is an extreme case and it may start with searching less than 10 knode per second only to jump later to hundreds of knodes per second In the end of this post there is an example when I told Deep Fritz 64 mbytes hash Can you explain me what happen and what is the reason that Fritz's nodes per second jumped? I did not run new applications during the run of Fritz so it is logical to expect that the number of nodes per second will be constant but it does not happen. Note that with p850(windows98) I have not similiar problems but I still found that the number of nodes per second of Gandalf was not constant and in this case it was a special problem of gandalf4.32 as a chessbase engine. Sphinx Dominator - MChess 486 r5k1/Bp3p1p/5bp1/1N3b2/2r1p1n1/6P1/P3PPBP/R2R2K1 w - - 0 1 Analysis by Deep Fritz: 21.Bd4 Rc2 22.Bxf6 ³ (-0.50) Depth: 2/7 00:00:00 21.Bd4 Rc2 22.Bxf6 ³ (-0.50) Depth: 2/7 00:00:00 21.Bd4 Rc2 22.Bxf6 ³ (-0.50) Depth: 2/7 00:00:00 21.Bd4 Rc2 22.Bxf6 ³ (-0.50) Depth: 2/7 00:00:00 21.Bd4 Rc2 22.Bxf6 ³ (-0.50) Depth: 2/7 00:00:00 21.Rac1! ³ (-0.47) Depth: 2/10 00:00:00 21.Rac1! Rxc1 ³ (-0.44) Depth: 2/10 00:00:00 21.Rac1 Rxc1 22.Rxc1 b6 23.Bxb6 ³ (-0.62) Depth: 3/13 00:00:00 21.Bd4! ³ (-0.59) Depth: 3/13 00:00:00 1kN 21.Bd4! Rc2 ³ (-0.50) Depth: 3/17 00:00:00 1kN 21.Bd4 Rc2 22.h3 Ne5 ³ (-0.44) Depth: 4/13 00:00:00 2kN 21.Bd4 Rc2 22.h3 Ne5 23.g4 ³ (-0.50) Depth: 5/17 00:00:01 10kN 21.Bd4 Rc2 22.h3 Ne5 23.g4 Be6 24.Bxe4 = (-0.22) Depth: 6/19 00:00:04 30kN 21.Bd4 Rc2 22.h3 Ne5 23.g4 Be6 24.Bxe4 Rxe2 = (-0.22) Depth: 7/21 00:00:10 101kN 21.Bd4 Ne5 22.h3 h5 23.Kh2 Ra6 24.f3 exf3 25.exf3 = (-0.12) Depth: 8/25 00:00:21 529kN 21.Bd4 Ne5 22.f4 Ng4 23.Bxf6 Nxf6 24.Nd4 h5 25.Nxf5 gxf5 26.Kf1 = (-0.06) Depth: 9/26 00:00:23 2085kN 21.Bd4 Ra5 22.Bxf6 Nxf6 23.Nd4 h6 24.Nxf5 Rxf5 25.Rab1 Ra5 26.Rb6 Kg7 = (-0.06) Depth: 10/27 00:00:27 4822kN (Blass, Tel-aviv 13.01.2002) Uri
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.