Analyzing Your Games with Supercomputer • page 1/1 • General Chess Discussion • lichess.org

I wonder is that any differences (huge difference) if we analyze our game with supercomputer than with a standard computer? Anyone does that?

MentalFugues

#2

There would be differences. The real question is whether most (if not any) of us could make use of the differences. Probably not. You're far better off getting advice from an expert human who can provide concepts and suggestions you can understand and effectively apply. Even if that feedback is theoretically less than accurate at the highest achievable levels for computers, it can be far more useful in games against your fellow humans

mkubecek

#3

Unless you also use a completely different engine, the only difference would be that you would get to the same depth faster (or deeper in the same time).

That being said, IMHO even current engines run on reasonable desktop are already too strong to be useful for practical evaluation purpose. One thing to realize is that if you had a perfect engine that would objectively evaluate a position assuming perfect play from both sides, it would only produce one of three possible evaluations: white winning, black winning, draw. And I feel today's engines are already a bit too far in that direction. They tell you "0.0" whether it's a dull endgame where you just need to push pieces without an obvious blunder or a sharp tactical position where only one narrow path leads to a draw and even top GMs would have hard time finding it. When an older engine said "+3.0", it was an equivalent of being a piece up and any decent player should be able to convert. Today the engine says "+4.2" and all I can say is that, yes, white seems a bit better but I'm not sure at all it would be enough. (Or, worse, white may have a pawn and active position for an exchange and if I played it, 15 moves later I would be still exchange down but without any compensation and desperately trying to save at least a draw.)

Unless you also use a completely different engine, the only difference would be that you would get to the same depth faster (or deeper in the same time). That being said, IMHO even current engines run on reasonable desktop are already too strong to be useful for practical evaluation purpose. One thing to realize is that if you had a perfect engine that would objectively evaluate a position assuming perfect play from both sides, it would only produce one of three possible evaluations: white winning, black winning, draw. And I feel today's engines are already a bit too far in that direction. They tell you "0.0" whether it's a dull endgame where you just need to push pieces without an obvious blunder or a sharp tactical position where only one narrow path leads to a draw and even top GMs would have hard time finding it. When an older engine said "+3.0", it was an equivalent of being a piece up and any decent player should be able to convert. Today the engine says "+4.2" and all I can say is that, yes, white seems a bit better but I'm not sure at all it would be enough. (Or, worse, white may have a pawn and active position for an exchange and if I played it, 15 moves later I would be still exchange down but without any compensation and desperately trying to save at least a draw.)

mvhk

#4

@agungss said in #1:

I wonder is that any differences (huge difference) if we analyze our game with supercomputer than with a standard computer? Anyone does that?
I know a 2600 fide player using a supercomputer. It is not cheap.
Difference is speed. For one second on the supercomputer you need one minute on a standard computer. Speed becomes important if you want to study thousands of positions in little time.

@agungss said in #1: > I wonder is that any differences (huge difference) if we analyze our game with supercomputer than with a standard computer? Anyone does that? I know a 2600 fide player using a supercomputer. It is not cheap. Difference is speed. For one second on the supercomputer you need one minute on a standard computer. Speed becomes important if you want to study thousands of positions in little time.

kajalmaya

#5

I think at lower levels if SF 15 running on my old laptop says move X is the best and a super computer running SF 17 says move Y is the best, I still do not know how I would make use of this difference in information. For a super GM maybe the difference is useful to get a different perspective on a position. So the bottleneck is how much of the analysis I am able to absorb, and that is very little.

For ordinary analysis where I may have made a mistake and computer (even old laptop running older version of Stockfish) points out a much better move, it is useful in that I know I made a mistake, and more useful if I really study the computer variation and its sidelines.

Another point is increasing the depth of analysis requires exponentially more resource, even if the same depth on a powerful computer would be achieved quicker.

I think at lower levels if SF 15 running on my old laptop says move X is the best and a super computer running SF 17 says move Y is the best, I still do not know how I would make use of this difference in information. For a super GM maybe the difference is useful to get a different perspective on a position. So the bottleneck is how much of the analysis I am able to absorb, and that is very little. For ordinary analysis where I may have made a mistake and computer (even old laptop running older version of Stockfish) points out a much better move, it is useful in that I know I made a mistake, and more useful if I really study the computer variation and its sidelines. Another point is increasing the depth of analysis requires exponentially more resource, even if the same depth on a powerful computer would be achieved quicker.

mkubecek

#6

@kajalmaya said in #5:

I think at lower levels if SF 15 running on my old laptop says move X is the best and a super computer running SF 17 says move Y is the best, I still do not know how I would make use of this difference in information.
Depends on what "best" actually means in each case. Sometimes this happens because deeper analysis is needed to recognize that a move is in fact good/bad and the evaluations differ a lot between shallow and deep. But these cases are quite rare, more often it's like X being +0.8 and Y +0.9 at depth 20 while X is +0.98 and Y +0.96 at depth 40 (and at level 50, Y might be "better" again).

@kajalmaya said in #5: > I think at lower levels if SF 15 running on my old laptop says move X is the best and a super computer running SF 17 says move Y is the best, I still do not know how I would make use of this difference in information. Depends on what "best" actually means in each case. Sometimes this happens because deeper analysis is needed to recognize that a move is in fact good/bad and the evaluations differ a lot between shallow and deep. But these cases are quite rare, more often it's like X being +0.8 and Y +0.9 at depth 20 while X is +0.98 and Y +0.96 at depth 40 (and at level 50, Y might be "better" again).

CM Sarg0n

#7

Most errors will be detected by your dishwasher...

lizani edited

#8

The strongest engines are the chessplayer's version of god and they are supposed to know the absolute truth in any position. However, you rarely need analysis from a 3000+ rated engine. Someone rated (x + 200) where x = your rating is sufficiently strong to give you a decent analysis which is easier for a human to understand than engine generated variations.
In live broadcasts, people often ask 'What does sesse or leela or some other high powered engine think?' To me, these are pointless questions as the default stockfish in lichess already gives you a pretty accurate view. If you want more certainty, you just have to use your brain a little and do some analysis yourself.

The strongest engines are the chessplayer's version of god and they are supposed to know the absolute truth in any position. However, you rarely need analysis from a 3000+ rated engine. Someone rated (x + 200) where x = your rating is sufficiently strong to give you a decent analysis which is easier for a human to understand than engine generated variations. In live broadcasts, people often ask 'What does sesse or leela or some other high powered engine think?' To me, these are pointless questions as the default stockfish in lichess already gives you a pretty accurate view. If you want more certainty, you just have to use your brain a little and do some analysis yourself.

kindaspongey edited

#9

@Journeyman2024 said in #57:
(https://lichess.org/forum/general-chess-discussion/are-most-casual-chess-players-low-iqtalentlesslazylow-effort)

... I don't postmortem my games because I don't know how, and I am 100% against artificial intelligence analyzing my games. Yet I don't see coaches on here offering even so much as a crash course on how to analyze your own games, which in turn is stopping those queries dead cold. ...
"... No matter which aids are available, you should first review the game, preferably with your opponent, and especially if the game was played at a decently slow time limit and your opponent is willing! ... If your opponent is not available to review the game, then you should still do so yourself at least once before showing it to stronger players or submitting it for computer analysis. ..." - NM Dan Heisman (2005)
https://web.archive.org/web/20140627023809/http://www.chesscafe.com/text/heisman50.pdf
https://lichess.org/forum/general-chess-discussion/how-to-analyze-our-own-games
Looking at the
https://lichess.org/Sud4qxCI

game, after 1 e4 c6 2 d3 e6 3 Nf3 g6 4 Be2 Bh6 5 Nbd2 d5 6 O-O Nf6 7 g3 O-O 8 h4 a5, isn’t it pretty clear, even without the computer, that 9 d4 was a mistake? Looking at the position after 8...a5 now, do you see a way that you could have anticipated that 9 d4 would probably not be a good idea?

@Journeyman2024 said in #57: (https://lichess.org/forum/general-chess-discussion/are-most-casual-chess-players-low-iqtalentlesslazylow-effort) > ... I don't postmortem my games because I don't know how, and I am 100% against artificial intelligence analyzing my games. Yet I don't see coaches on here offering even so much as a crash course on how to analyze your own games, which in turn is stopping those queries dead cold. ... "... No matter which aids are available, you should first review the game, preferably with your opponent, and especially if the game was played at a decently slow time limit and your opponent is willing! ... If your opponent is not available to review the game, then you should still do so yourself at least once before showing it to stronger players or submitting it for computer analysis. ..." - NM Dan Heisman (2005) https://web.archive.org/web/20140627023809/http://www.chesscafe.com/text/heisman50.pdf https://lichess.org/forum/general-chess-discussion/how-to-analyze-our-own-games Looking at the https://lichess.org/Sud4qxCI game, after 1 e4 c6 2 d3 e6 3 Nf3 g6 4 Be2 Bh6 5 Nbd2 d5 6 O-O Nf6 7 g3 O-O 8 h4 a5, isn’t it pretty clear, even without the computer, that 9 d4 was a mistake? Looking at the position after 8...a5 now, do you see a way that you could have anticipated that 9 d4 would probably not be a good idea?

Journeyman2024

#10

That's one of my problems in a nutshell, because in all honesty, no, I cannot. I'm not neurotypical. A plain English explanation of why 9. d4 was a mistake would be needed for me to see it, and I'm...not seeing it.

About all I can say for myself is that it would behoove me to ask a question on the forums about how to improve my game knowing this, but I have no clue how to word it without sounding whiny (which would not be intention, but it's not that easy to interpret intention from typed words).

That's one of my problems in a nutshell, because in all honesty, no, I cannot. I'm not neurotypical. A plain English explanation of why 9. d4 was a mistake would be needed for me to see it, and I'm...not seeing it. About all I can say for myself is that it would behoove me to ask a question on the forums about how to improve my game knowing this, but I have no clue how to word it without sounding whiny (which would not be intention, but it's not that easy to interpret intention from typed words).