Sometimes a very long time. IMO it's always good to input your ideas first. Sometimes it takes a few moves before the program finally agrees. Sometimes it never agrees (because the move it suggests is based on deep tactical calculations).
And the computer missing the best move happens more often than you might think. Given a long time to think they basically never blunder, but that doesn't mean the move it picks is best... otherwise ICCF wouldn't be a thing.
so then, exactly, how reliable is engine analysis and does it matter at our level?
i was just analyzing an endgame i played with Fritz. and it no doubt picks better moves than me and considers variations i didn't dream of.
Sometimes engines need minutes to find the correct continuation.
I find it amusing when chess engines failed to find the winning moves in some positions, chess engines are being doubted for their inability. As if chess engines have to be perfect to be trusted as analysis tools. I believe top chess engines can find the best moves 95% of the time in less than 2 minutes in a quadcore PC.
If you use houdini 3 then you can set the option "Tactical mode" (it is off by default).
This mode makes the engine find combinations faster but use it only for sharp positions as it otherwise decreases positional strength of houdini.
Engines suffer from so called "horizon effect" in some positions. They sometimes underevaluate positions arising after 20-30 plies (10-15 moves) which makes them consider another lines better. If a combination leads to a winning endgame an engine needs sometimes to do much deeper calculations to see that endgame winning indeed (experienced players can evaluate the result of an endgame based on some general principles but enggines have to calculate them unless there are at most 6 men on the board and they can use Nalimov).
I had several similar situations in my games when the engine had to go for depth over 40 to choose my move and reaching such depths is slow even on a modern quadcore CPU.
I see much talk of multi-core CPUs x GB hash, parallel comuting, running on network of computers, 32-bit vs 64-bit OS etc..Is any of this relevant. pdve
Hell yes it's relevant! But not to you or me, for all practical purposes - unless you really think Houdin 3 (ELO 3300) is going to make you any better than some dink engine with a crappy ELO rating of 2800...
SOURCE: http://chess-table.blogspot.com/2012/12/download-best-chess-engines.html
Top Free Chess Engines running on a single CPU in a 32-bit OS
http://computerchess.org.uk/ccrl/4040/cgi/compare_engines.cgi?family=Fritz&prin
| 1 | Deep Fritz 13 4CPU | 3052 | +19 | −18 | 41.4% | +51.0 | 51.4% | 868 | |
|
99.8%
|
|||||||||
| Deep Fritz 11 4CPU | 2999 | +29 | −29 | 54.2% | −27.2 | 55.2% | 335 | ||
|
57.4%
|
|||||||||
| Deep Fritz 12 4CPU | 2995 | +34 | −34 | 46.7% | +22.5 | 52.5% | 255 | ||
|
92.0%
|
|||||||||
| Fritz 13 | 2969 | +11 | −11 | 45.7% | +25.5 | 49.3% | 2532 | ||
|
100.0%
|
|||||||||
| Deep Fritz 10.1 4CPU | 2894 | +18 | −18 | 51.9% | −13.4 | 39.1% | 1048 | ||
|
69.0%
|
|||||||||
| Fritz 12 | 2889 | +11 | −11 | 50.8% | −5.0 | 48.3% | 2838 | ||
|
100.0%
|
|||||||||
| Fritz 11 | 2858 | +10 | −10 | 52.7% | −18.6 | 46.5% | 3608 | ||
|
62.7%
|
|||||||||
| Deep Fritz 10 2CPU | 2852 | +37 | −37 | 60.8% | −66.9 | 44.9% | 227 | ||
|
81.7%
|
|||||||||
| Deep Fritz 10 4CPU | 2832 | +20 | −20 | 49.0% | +6.8 | 40.8% | 826 | ||
|
100.0%
|
|||||||||
| Fritz 10 | 2780 | +15 | −15 | 51.7% | −11.6 | 37.3% | 1489 | ||
|
100.0%
|
|||||||||
| Fritz 9 | 2743 | +13 | −13 | 50.1% | −0.9 | 34.7% | 2105 | ||
|
79.4%
|
|||||||||
| Deep Fritz 8 2CPU | 2733 | +19 | −19 | 50.8% | −4.3 | 34.1% | 909 | ||
|
97.5%
|
|||||||||
| Fritz 8 Bilbao | 2700 | +26 | −26 | 45.8% | +31.0 | 36.0% | 483 | ||
Perhaps slightly higher 97% - some great composed chess studies (not involving zugzwang) are unsolvable from start position.
Try past this FEN: kb6/p5P1/P6p/1K3P1P/3p3P/3P4/1p1N4/qB6 w - - 0 4
Check if your primary engine of choise can spot the bishop promotion.
so 3300 elo is approximately 2.5 times my rating. i have ordered rybka from the chessok shop so it doesn't seem to me that i will notice any difference between 3100 and 3300. anyway my idea is to use the interactive deep analysis feature to analyze my pet lines in the french and sicilian.
also, i notice that analyzing lines from both the white and black side can lead to confusion so it is best to analyze only the lines you play as either black or white. aquarium also has a feature where you can set a 'preferred side'. this is why i am splurging my money on this.
also, i notice that analyzing lines from both the white and black side can lead to confusion so it is best to analyze only the lines you play as either black or white. aquarium also has a feature where you can set a 'preferred side'. this is why i am splurging my money on this.
It never confused me and Fritz also has options to analyze from either side or both. I'm willing to bet right now without ever having seen any other chess program in several yrs they all (not counting freeware) have this option Of course maybe aquarium is the best program for you, but don't base decisions like this on facts you just made up.
I see much talk of multi-core CPUs x GB hash, parallel comuting, running on network of computers, 32-bit vs 64-bit OS etc..
Is any of this relevant. I normally run my Fritz 13 and pick the best move it gives in the first few seconds which is normally quite good.
However, my faith in chess software was shaken recently when i inputted a position which according to my human intuition was winnable by a double sacrifice at h7. Fritz was unable to find it. Houdini 2 was unable to find it. When I actually inputted the sacrifice, fritz and houdini agreed it was winning. So what is the deal? How long should I run these programs before their results match reality?