Author: Uri Blass
Date: 15:40:04 03/07/04
Go up one level in this thread
On March 07, 2004 at 17:43:10, Chessfun wrote: >On March 07, 2004 at 14:22:50, Uri Blass wrote: > >>On March 07, 2004 at 11:46:26, Chessfun wrote: >> >>> >>>Minimum 70 games to qualify. >>>All 3,4 and 5 piece tablebases. >>>Native books where made, general.ctg where not. >>>Majority 9100+ games over 60 mins. >>>Games are autoplayed ponder=on and single PC ponder=off. >>>AMD XP 2600+ >>> >>>Hiarcs8 Bareev games at 60' ponder=off using Hiarcs9.ctg >>> >>>Currently testing Ruffian 2.1 with Lokasoft book. >> >> >>Is movei the next program to be tested? >> >>I sent to Leo a new version that is probably 1-2 elo better thanks to being >>faster unless I have an unexpected bug. >> >>It is without logfiles that you cannot disable and in order to have logfile with >>the new version you need to enable logfile. >> >>I hope that it has not new bugs. >>I tested that it get the same number of nodes in some test positions faster. > > >Ok Uri since I'm currently running two pc's to catch up, tell you what. You tell >me the hash size you want to use and I'll give it a whirl. > >"I believe that 64 mbytes are better than 16 mbytes at 1 minute per game but not >at 4 minutes per game because at 4 minutes per game movei clears hash after >every move when it is not in time trouble so it spends time about it relative to >the case of 16 mbytes." > >"It is possible that at longer time control the time of clearing hash is not >significant and 64 mbytes becomes again better than 16 mbytes." > >The games will be at 30'/40+30'/40+15' > >Sarah. It is an hard problem because the fact that I am the programmer does not help me to know the best hash size and I did not spend computer time to investigate it. If people can give me results of test suites at 2 minute per move with different hash size then it may be productive for me(I simply did not have a lot of computer time for my program and one of the reason that I release my program is also that I hope that people will help me by giving me information). 174 was tested at 90 minutes against 30 minutes(no pondering) under Fritz with 16 mbytes and there was no problem with it when the hardware is A1000Mhz. 178 was not tested in games and I hope that there is no problem with it. 178 is probably only slighly faster than 174 that makes it maybe 2 elo better I do not think that 16 mbytes or 64 mbytes is very important at your time control but I did not test latest versions with both sizes. I think that time of clearing hash tables is not very important at 30/40 so the main question is if movei performs better with more hash and results of test suites can be productive to help me(one test suite is not enough because of luck problem and I probably need many hundreds of positions that means many hours of computer time). Movei at your time control will usually clear the hash size both with 64 mbytes and 16 mbytes. If the remaining time is too small it does not clear the hash but too small happens rarely and can happen in your case only in the last moves of the time control even with 64 mbytes. 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.