Author: Christophe Theron
Date: 00:19:10 05/17/99
Go up one level in this thread
On May 16, 1999 at 22:26:11, James Robertson wrote:
>On May 15, 1999 at 20:36:14, Christophe Theron wrote:
>
>>On May 15, 1999 at 18:42:14, Roger D Davis wrote:
>>
>>>I tried to run a tournament with Little Golith Gold 2 last night using the
>>>Winboard adapter in Fritz 5.32. Had no problem downloading Little Goliath and
>>>getting it to appear as an engine in Fritz, but the results were rather
>>>disappointing given what I've heard about the engine, so I'm not sure whether I
>>>did things correctly.
>>>
>>>Anyway, I rebooted by machine and set up a tournament with Junior 5, Fritz 5.32,
>>>and LGGold 2. It was "Gauntlet" tournament, since I didn't care to let Junior
>>>play Fritz. I edited the .ini file for LGGold to give it 52 megs of hash (my
>>>machine is a PII 350 with 256 megs RAM). When I started the tournament, I had to
>>>choose a hash size for all programs, so I chose 64 for Junior and Fritz, and 64
>>>for LGGold. I don't know if this setting overrides the .ini file or not. Perhaps
>>>it does and 64 is worse than 52 (a recommended value in the read.me file). All
>>>programs used general.ctg as book. Time controls were 10 minutes per game, no
>>>increment.
>>>
>>>I watched the first two games and went to bed. In the first, Junior crushed
>>>LGGold with a sacrifice. In the second, versus Fritz, LGGold kept evaluating a
>>>drawn position as if it were a pawn ahead, while the evaluation by Fritz was
>>>0.00.
>>>
>>>What was most disturbing, however, was that LGGold was getting 35-40K nodes per
>>>second just out of book, while Fritz and Junior were getting 250-300K nodes. I
>>>thought that since LGGold was supposed to be this tactical monster, it would
>>>search more nodes per second.
>>
>>Nodes per second and tactical abilities are not related to each other. I know
>>it's a well spread opinion that fast searchers are good in tactics but there is
>>no objective/scientifical reason to think so.
>
>40k NPS is odd because LGG generally gets NPS similar to Fritz or Junior (at
>least that is the standard report).
>
>James
I'm not denying this.
I deny the assumption that: "high NPS <=> program good in tactics".
It is possible to write a tactical monster that has a low NPS (but I don't imply
that such a program would be very strong).
Christophe
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.