Author: Peter Kasinski
Date: 13:10:20 10/26/05
Go up one level in this thread
On October 26, 2005 at 12:22:31, Uri Blass wrote: >On October 26, 2005 at 11:30:44, Sedat wrote: > >>On October 26, 2005 at 11:09:20, Peter Kasinski wrote: >> >>>- Match of 24 games >>>- Fruit 2.2: FX-55 2.6 GHz; 512 Mb hash >>>- Deep Junior 9: Athlon 64 X2 4600+ 2.4GHz Dual Core; 1408Mb hash >>>- Fruit using Shredder GUI and its own book >>>- Junior with 3,4,5 and 138Gb of 6-man tablebases >>>- Time Control: 40 in 120, 20 in 60, rest in 30 >>> >>>Game 14: Fruit wins >>>Game 15: draw >>>Game 16: Junior wins (first of the match) >>> >>>Fruit leads 11.0 - 5.0 (+7 -1 =8) and needs 1.5 points from 8 to take the match. >> >> >>Hello Peter, >> >>Thanks a lot for the update... >> >>I love to join and watch such interesting matches on super fast computers !! >> >>I see that Deep Junior 9 uses both cores and even this processor addvantage does >>not help to DJ9 !! >> >>Best, >>Sedat > >I am not sure if the use of 1408 Mbytes is an advantage for Junior. > >Note that programs usually are not tested in these conditions and >I doubt if Amir has so much hash to use in his tests so it is possible that >Junior is weaker from too much hash tables. >I remember that Junior blundered by avoiding Rxe6 in one of the games and it is >possible that using too much hash was one of the reasons for that. > >Uri Junior's middlegame NPS is typically 3 to 4 million on this hardware. During WCC it ran on a 4-way Opteron at (presumably) twice the speed. Given that, the suggested detrimental effect of large hash tables would seem like an inexcusable deficiency to me. Of course, this is just a conjecture. The unfortunate 31.Qe5? could have been caused by another problem altogether. PK PS. Come to think of it: would you happen to know what Amir's tournament setting for Junior's hash table size would be?
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.