the great blessing of chessgames.com is that it affords users, subscribers and lurkers alike, the digital "ignore" feature.

the great blessing of being human is that it affords you, regardless of circumstances, the personal "ignore" feature.

for my part, and in a helpful spirit, i find the "dump" of computer analysis (yours and those by <mel gibson>) to be completely unhelpful.

not only are the defensive/offensive alternatives omitted, which defeats any pedagogic purpose, they are just hard to parse. (some are even posted without move numbers.) and pursuing the analysis of a 23 move game into move 43 or 52 seems to me vacuously factitious.

i'm not spurning silicon at all; i play stockfish on my ipad, i thrilled to following the alpha zero games, every one of them. it's really enlightening to see how legends like reti and spielmann stand up to modern analysis.

I would say that it depends on what you are searching for. If you are looking for beautiful victories that are often the result of human mistakes made during the game then, yes, computer game-playing and computer analysis of games <may> be "destroyed". As far as the latter I fail to see much difference between a post-game computer analysis and a post-game human analysis conducted with much more time than was available during the playing of the game, with the benefit of a chessboard where the pieces can be moved, and a word processor or notebook where the game's analysis and notes about the game can be written. The only difference I see is that the human analysis, unless supported by <well done and vetted> computer analysis, is likely to contain more errors.

But perhaps what you enjoy are the likely human errors during a game as a result of time control pressure, position visualization and human calculation limits, and any other external non-game-related circumstances; e.g. what was the player's frame of mind at the time that the game was played. Fair enough. In that case you probably prefer games between humans played as the faster time controls, Rapid, Blitz, or even Bullet since they will likely have more errors than games between humans played at Standard/Classic time controls, not to mention "Correspondence" time controls. Again, fair enough. We all have our preferences.

FWIW, although I definitely enjoy seeing flashy and even unsound attacks (my all-time favorite player is Tal) my preference is to determine the "best" moves available after each move by an opponent. And note that finding the "best" moves (a.k.a. "truth") even with computer assistance is likely an unattainable goal since computer analysis must be vetted to uncover possible better lines that were inadvertedly pruned from the engine's search tree and adverse results of the horizon effect, including the current inability of chess engines to always recognize fortress-type position. But De Gustibus Non Disputandum Est.

And the quality of computer analysis posted here and probably everywhere else is of variable quality. Did the analyst run the analysis deep enough? Did the analyst take into account the non-determinism of chess engines running on multicore systems (run the same analysis on the same computer at different times and you will get different results)? Did the analyst use different engines (different engines will give different evaluations of possible moves and may differ in their move rankings)? Did the analyst properly vet the computer's suggested lines? And this is not an exhaustive list.

So don't necessarily assume that your comments are necessarily substituted by a computer. Depending on the skill and time that the analyst running the computer analysis have available, his/her analysis might not be as good as yours. Don't give up, your kibitzing is still wanted.

You say you moved to music long ago? In that case you are probably just as unhappy with the state of computer music composition as you are with the state of computer chess playing and analysis. Have you looked at what is being done by computers with regard to music? Take, for example, https://www.youtube.com/watch?v=CsQ... or https://www.youtube.com/watch?v=f_o.... You can look up the tool used, computer.bot, if you like.

Also note the many works by David Cope; e.g. https://www.youtube.com/watch?v=2ku..., https://www.youtube.com/watch?v=o7z..., and https://www.youtube.com/watch?v=Pcz.... If you (or Vivaldi, Mozart, and Bach respectively) can do better, my hat's off to you and them.

Since I am interested in improving those computer-based "analyses" that I post, could you please elaborate? Here are some examples of some of the computer analyses I have posted relatively recently:

A Zapolskis vs S Maze, 2016 (kibitz #13)

Schlechter vs Tarrasch, 1911 (kibitz #18)

And this is a 3-part analysis:

D E Cori Tello vs A Travkina, 2010 (kibitz #8)

D E Cori Tello vs A Travkina, 2010 (kibitz #9)

D E Cori Tello vs A Travkina, 2010 (kibitz #10)

Having said that, another school of thought considers analysis as 'incomplete' unless you dwell on a move for a period of time and actually determine what the 'next best' candidate moves are, listing them, and then rejecting them with short explanations of best or 'typical' play. You have to figure out how certain you want to be, and then make your text reflect that. Just my two cents. Consider my first post in this game, a 24-minute dwell on move 16, where I declare that 17...Rf7 had 'defensive chances.' I did not know how the position would turn out, and some may not have cared. I later listed a 'critical line' of sorts and a 'critical position,' which I then examined in depth for hours. This type of analysis is accepted by some but not by others. You can choose how paranoid you want to be in ignoring inconsequential moves, but there will always be those who disagree with some shortcut.

In your next game you went through the final positions with a 7-piece tablebase and made observations. That was clever and might have added value and understanding to someone without access.

well, to my eye, your analyses are well "curated" (to use the trendy term of the day) and the positional snapshots are very helpful. your posts are clearly not "cut and paste," and you puzzle over steps along the way. i find these helpful.

my main complaint with computer analyses is that they are linear. chess is inherently branching; the computer analyses prune every tree to a stick.

i especially notice kibitzing that clarifies tactical faultlines. this is because i am a duffer, will die a duffer, but hope to die better than i am now.

i don't assume computers give definitive answers. along comes a better computer, has a different answer, beats the old computer in a tournament.

I do think it is important to remember that if the position is ‘even’ for a 3300 rated computer, but requires 10 perfect moves to defend, it is likely a good move to play against a human.

click for larger view

Stockfish_20080713_x64_modern:

41/84 10:18 -2.18 16.Qg5 c6 17.Bd3 Nxd5 18.Qxd8 Rxd8 19.Bxf5

41/69 10:18 -2.72 16.Nxf5+ Bxf5 17.Rxf5 Re8 18.Qc1 Rxe1+ 19.Qxe1

What program do you use to run Stockfish?>

Thanks for your comment.

The issue is then: How deep must we let a given engine analyze in order to have a reasonably high confidence that the engine's evaluations are reasonably correct? That depends on the engine. Stockfish, thanks to its very aggressive search tree pruning, can reach deeper depths than other top engines in the same amount of time. But, again due to its very aggressive search tree pruning, it misses more promising lines (search tree branches) at lower depths than other top engines. So it must search deeper in order for one to have equivalent confidence in its move rankings (the value of its evaluations are not important with regards to its game playing prowess).

So I now think that in order to determine how deep the engine needs to search in order to sufficiently accurately determine the best move to play (in reality what the principal variation is) in any given position we must analyze the pattern of its move rankings. I'm sure that you have seen many analyses where one move is considered to be the best (i.e. has the highest evaluation) at low search depths and the difference between its (changing) evaluation and the (also changing) evaluation of the next best move increases with search depth. Conversely, sometimes a particular move will have a large difference during the early search plies in its evaluation compared to the second best move but that difference keeps getting smaller the deeper the engine searches and the implication is that at some greater depth the rankings of the two moves will be reversed. But in both cases we must be on the lookout for the AKC2L (AylerKupp's Corollary to Murphy's Law) effect which you can find defined in my forum's header.

In the first case we can probably stop the analysis with a large degree of confidence that the engine has found the best move. In the second case, depending on the magnitude of the difference, we could either select the second best move as (eventually) becoming the best or let the analysis run longer to see if that really happens.

But then there are situations when the difference between the evaluation of several moves is relatively small, and the move rankings change every few search plies or even after every ply. In those cases I think that we have to accept that the engine can't accurately determine what the best move in that position really is, and that the selection of the "best" move depends on when you stop the analysis. You just have to select as the "best" move the move that looks best to you, with all the subjectivity that might imply.

And, of course, different engines will make different evaluations of the candidate moves and their resulting lines, and even the same engine will do the same as a result of the non-determinism of multi-core engines. So there will always be uncertainty in the level of confidence we can attach to a chess engine analysis, even if we run the analysis using different engines and/or run the analysis with the same engine multiple times.

But computer analyses are also inherently branching. They probably calculate and evaluate many, many more move and line possibilities for a single position than a human could calculate in a lifetime. And eventually both humans and computers must prune every tree to a single stick since neither of them can play more than one move from a given position.

<i don't assume computers give definitive answers. along comes a better computer, has a different answer, beats the old computer in a tournament.>

I'm assuming that when you say "computer" you mean a combination of hardware and software. You can run the same software (chess engine) on two computers, one much more powerful than the other, and most likely the computer with the same software but much more powerful hardware will have the (much) better scoring percentage.

Or you can run the different software on the same hardware. Then the better software will most likely have the (much) better scoring percentage.

Or to make it more "interesting" you could run the less capable software on the more powerful hardware and try to see what combination of hardware + software will provide better results. You can also make it even more "interesting" by giving one computer (the one with less powerful hardware and less capable software) more time than the other computer and see what kind of time odds are needed to allow the weaker computer to achieve an equal scoring percentage than the stronger computer. At least that's interesting to me and a great way to waste a lot of time.

<i don't intend to impinge on anyone's enjoyment or preferences. your business is none of mine and, as i said, it's just a game.>

I have some suggestions for you. First, you only really have 4 physical cores in your computer, the other 4 "cores" are really "phantom" cores will lesser capabilities as a result of Intel's hyperthreading even though they are reported by Windows as being "CPUs". Most chess engine developers suggest not using hyperthreading when running their engines and some even indicate that the resulting effect of using hyperthreading is to have <less> effective computing power available for use by the engine because of the overhead of switching between threads.

Second, there is nothing wrong with your system running at 100% CPU capacity for prolonged amounts of time (e.g. for an overnight analysis) as long as it doesn't overheat, something that might happen on a laptop with limited cooling. Yes, you might notice delayed response times, sometimes much delayed, if you are running your engine specifying 4 threads while you are attempting to run another program like a word processor. With Arena you can solve this problem by assigning Arena a much lower priority than other programs by selecting 'Engines' from the main menu and then 'Manage' > 'Options' > 'Low priority' from the pop-up menu. Afterwards, whenever you switch to your word processor (such as hitting a key on the keyboard) the necessary computing resources (a computer core, memory, etc.) will be allocated to your word processor and you will likely not experience much response time degradation.

True, Arena and Stockfish will run somewhat slower since one of its 4 cores will be dedicated to the other application but that would be temporary. And you might experience some response time degradation if your application has been swapped to disk and must be reloaded into memory in order to execute. But as long as you don't allocate more than 50% of your computer's RAM to the hash table there should be enough RAM available to hold the code for Arena, the chess engine, the other application, and whatever system process need to execute in memory so I suspect that any response time degradation will be small and infrequent.

As far Arena + the chess engine only running in Admin mode I can't say. It depends on what resources they require and whether your user account has sufficient privileges to run those programs. But as long as you don't try to change any system settings that you are not familiar with you should be able to run in Admin mode without any adverse effects.>

My archaic computer does not provide hyperthreading so I don't really know whether it speeds up or slows down Stockfish. You might experiment and see what your results say. But disabling hyperthreading would certainly help your overheating issues since there will be more times where portions of your core are not active. And if it also speeds up your Stockfish, that would be a win-win situation.

And win-win situations are rare. I was fortunate to experience one of them when I was designing missiles many years ago. The hardware designer responsible for the missile processor card had a choice of 2 processor chips, one which was clocked at 10 MHz and one at 12.5 MHz (I told you it was many years ago). The 10 MHz chip was cheaper and had a version that was Mil-qualified for harsh environments, something that would have been very expensive to do for the 12.5 MHz part. Still, since the hardware designer knew that software people always wanted more speed, he had based his design on the 12.5 MHz part.

But he had done timing analysis and thought that the margins for the processor accessing memory were a little bit tight and processor/memory access would not work reliably, particularly at higher temperatures. So he decided that he had to add a wait state to slow the processor down when accessing memory. His simulations indicated that this would work.

He was explaining this at a design review when I raised my hand and asked him if he had run his simulations with the 10 MHz part and he said he hadn't. I asked him to do that and when he did he found out that with the 10 MHz part he didn't have to include a wait state for memory access to have an adequate design margin. So, even though the processor was slower, the system as a whole was faster, the chip was cheaper, and it was Mil-qualified to boot.

Thanks Ayler,
I don't want to stuff around with this computer - an an i7 quad core 2600K. I'm looking to upgrade the one I have to a 16 core 32 thread Ryzen 3950X. I'm stuck in a Covid lockdown so I can't organise that yet.