Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Opening Books / Questions to speed, engines
- - By yesno [de] Date 2015-01-07 12:43
Hi all together!

I have several questions about Deep Rybka 4.

I use the following engines on Arena: Critter 1.6a, DeepHiarcs 14 wcsc, DeepShredder 12 x64, Houdini 4 x64, Stockfish 15010523 x64, Delfi, Gull 3 x64, Strelka 6 and of course Deep Rybka 4 x64.
Hash table size for all engines: 32MB, experimental with 512MB or such did not deliver any improvements imho.
All engines use their own books.
Nalimov cache: 64MB.
Gaviota tablebases 5.
System: 64bit Win7, 12GB RAM, Athlon II x4 640 processor.

1. All these engines, incl. the Stockfish hype stuff, lose most rounds against Deep Rybka and I am not sure how and why all those Houdinis and Shredders can win engine-competitions against Deep Rybka. Any ideas? (I use gaming mode "fixed search depth" in depth 8, 12, 16)

2. Why is Deep Rybka so slow in calculating? When Rybka searches for example 13 moves with 711k combinations, others like Stockfish or Houdini search 23 moves with 13 million combinations in the same time, but still they lose. Reasons?

3. Using the client "Arena", I always use the modus "fixed search depth", not the others like Blitz, mate search, endless, tournament or fixed time per move. The description of Arena says that using the fixed search depth (for example 16) is not really effective and does not call the best from a chess engine, but they dont say which of these modes can leech the best performance from an engine. Any idea which mode is the best to beat a human opponent?

Parent - By Dark_wizzie (***) [us] Date 2015-01-07 20:48
First, how many games did you do?
Second, what time control is it? Slower CPUs and shorter time controls means less hash is optimal.
Third, that's just the way Rybka rolls, the author made it show a low node count. Nodes can be counted in too many ways. You cannot compare node counts across engines to determine strength.
Fourth, why would you use fixed search depth? Just like node count, the depth count varies wildy from engine to engine. 30 depth Rybka is not at all fair to compare with 30 depth Stockfish. Instead maybe you should try giving them the same amount of time, because that's what really matters in the end. Try 3 minute games using the tournament feature. Or 5 minutes, or 10.
Fifth, the benefits of hash even when it's beneficial is generally not readily visible because the boost is not THAT large. When we're looking at this stuff sample size is important.
Parent - By user923005 (****) [us] Date 2015-01-08 09:28
1.  Don't use fixed search depth.  Programs calculate that differently, and Rybka  exaggerates {lies} about depth in plies.

Use time controls based on time rather than depth.

2.  It's a marketing ploy, meant to make the program look smart.

3.  The only reason I can think of to use fixed depth is to analyze a set of positions to a certain number of plies.  Each engine will count differently.  For instance, with Junior, single reply positions do not count as a normal ply, and Rybka does some monkey-business in ply reporting.

I never used fixed ply search for anything except analyzing EPD positions with constants appropriate for the engine.
Up Topic The Rybka Lounge / Opening Books / Questions to speed, engines

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill