Author: William Penn
Date: 07:53:40 12/11/04
Go up one level in this thread
On December 11, 2004 at 10:27:08, Derek Paquette wrote: >On December 11, 2004 at 09:51:34, William Penn wrote: > >>Shredder 8 tip: use Tablebase subsets >> >>There is far too much hard drive churning with Shredder 8 as simple endgames >>approach. The slowdown is enormous. The kN/s speed may drop to 10% of normal. I >>also fear for the life of my hard drive with such constant heavy churning. >> >>To return my system to sanity, I've developed some subsets of the tablebases >>which can generally be used with reasonable demands on my computer system. Of >>course the 3-4 piece TBs cause no problems, so can be used anytime. Only the >>bigger 5 piece TBs are causing the problem. As a rule of thumb selecting TBs of >>about 35MB or less is acceptable, and that comes to about 1GB total TBs. >>Selecting TBs of about 70MB or less is sometimes acceptable, and that comes to >>about 2.3GB total TBs. Of course the complete 3-4-5 piece TBs come to 7GB total, >>but my computer just can't handle them in simpler endgame positions. So there's >>no point to even considering the huge 6 piece TB files. >> >>FYI I'm running an XP 2400+ processor at 2GHz with 1GB RAM. >> >>It also helps to reduce the hash size. I normally run about 640-768GB hash, and >>reducing that to 256MB or 128MB improves the TB access situation. But sometimes >>that's just not enough in simpler endgame positions, so a smaller set of TBs >>must be used. >> >>This problem is far less severe with other engines. Shredder 8 is the worst, and >>the only engine I've tried that requires this kind of special compensation. I >>hope they will fix this in Shredder 9. >>WP > >Strange that it is the ONLY engine that does this, and I agree, although, >strange that shredder 8 dominates computer vs computer chess and that it has the >strongest endgame of all engines, > >coicidence? Thanks for your confirmation. It can be rationalized in that various studies indicate that tablebases have very little affect on the overall strength of an engine in ordinary chess play at the usual time controls. Of course it matters greatly to us when we want to analyze an endgame extensively to determine the truth, but that is not one of the criteria that goes into SSDF ratings. I wonder, however, what is your basis for saying that Shredder 8 plays the strongest endgame? Have any definitive tests been run by somebody? My experience doesn't agree... I'm convinced that there is a Shredder 8 bug related to handling of Rooks, for example 2R+pawns vs 2R+pawns in an open or semi-open & active position. It is fairly frequent that I see Shredder 8 simply failing to find even simple plans of action, and instead prefers to waggle its Rooks back & forth as if a draw were always acceptable. It doesn't know how to make progress, how to assume the initiative, how to break through a semi-closed position, etc. It also makes some dumb oversights in combinations when active Rook play is involved. I know, you & others want specific examples, and I tried giving a couple of examples here awhile back and got flamed as a result. So I'm not prepared to go to that much trouble - because it would do no good. Nothing gets fixed, even if there is good evidence, so why bother? I have a rule of thumb: Whenever active Rook play is involved, be sure to check the line with a different engine, because Shredder 8 is unreliable. I believe that. I get into trouble when I ignore it in my analysis. Of course I only use infinite analysis mode, analyzing each position for an hour or more. So these observations may not apply to "speed chess", i.e. anything faster than 1 hour per move... :) WP
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.