@drmrboss
I have the EGTB. I normally practise against it. But there are many slightly inaccurate moves leading to positions that still require a lot of thought from the opponent, but which you will practically never meet when practising random positions against an EGTB - hence the requirement for a slightly inaccurate program.
Programs are bad practice partners for the same reason... also because in positions that aren't easy to calculate, let's say some maneuvering rook endgame, sometimes they play outright blunders that a human wouldn't... but yes, they can play completely artificial winning or drawing schemes too.
Maybe some kind of "slightly inaccurate program" (as you put it) would work... but choosing stockfish or leela (they're known as the best) is a strange way of seeking out a "slightly inaccurate program" to practice technical endgames against.
I mean, maybe it would be good practice for you? I don't know. Just giving my impression of what you're saying.
Yeah like when the computer see that hanging it’s rook prevents mate in 43 and instead causes mate in 45. Something a human would never do, but technically the best move
@llama
SF8 is actually far too inaccurate. It just can't play positions with a depth of much more than 20.
(See e.g. https://www.chess.com/forum/view/endgames/computer-level-10-cant-beat-the-2-knights-vs-pawn-checkmate-drill)
@llama
SF8 is actually far too inaccurate. It just can't play positions with a depth of much more than 20.
(See e.g. https://www.chess.com/forum/view/endgames/computer-level-10-cant-beat-the-2-knights-vs-pawn-checkmate-drill)
I learned and practiced endgames like Q vs R and R+B vs R. Learning them once isn't so bad because there are good lessons in zugzwang and piece coordination, and you'll have a vague idea of what to do if you reach that position in a real game.
But you wont remember how to win them without practicing them from time to time... which seems awfully impractical for non pros. I'd put N+N+pawn in that category too.
I learned B+N mate that way. I would suggest sometime trying Q vs. N+N. It’s actually extremely hard
B+B vs N is also extremely hard.
I mean, again, I think 2600 GM is far too weak to bother being good at such endgames... and even Carlsen admitted he wasn't very good at first (compared to other top 10 players) in technical endgames like this.
They're just not very important to have mastered unless you're making a real push for the world champion title and it's your literal job.
I learned B+N mate that way. I would suggest sometime trying Q vs. N+N. It’s actually extremely hard
Yes I've tried that too. I can't play it yet. It's actually a prerequisite for two knights against pawn.
Interesting that GMs have long been agreed that it's generally drawn but the EGTB's think it's almost always a win for the queen.
...
They're just not very important to have mastered unless you're making a real push for the world champion title and it's your literal job.
I think not even then. Karpov reached a KNNKP endgame against Topalov in 2000 and made a complete pigs ear of it. But he didn't do badly on the whole.
Those endgames are the real spirit of chess!
Meh.
The middlegame and calculation is the real spirit of chess.
But sure, understanding endgames is important... not so much those endgames, but yeah, knowing where the drawing margins are is important. It helps you choose even better middlegame ideas 
I already did that, and posted the max and min win / draw for an 18000 game match, and tried to explain it to people...
Even so, the Elo formula doesn't predict real results when the players are that far apart. Specifically, we know from practice that the lower rated player scores better than predicted.
But yes, looking at the expected score will let you know the max and min win and draw percentages... there is a range because the formula doesn't distinguish between a win and two draws e.g. for a 6 game match, win/loss/draw, a record of 4/2/0 is the same as 2/0/4
No, a 3000 would defend successfully
According to the following formula: 
If player A is 4000, and player B is 3000, the expected score of player A would be: 1/(1+10^((3000-4000)/400)), as 1noobchesser said, this gives roughly 99% win rate. Same goes for 3000 vs 2000. So, theoretically a 4000 vs 3000 should score the same as a 3000 vs 2000. I'm pretty sure people already mentioned all of this. But the real question is whether something can achieve 4000, this brings us to another: how close are we to perfecting chess?
Meh.
The middlegame and calculation is the real spirit of chess.
But sure, understanding endgames is important... not so much those endgames, but yeah, knowing where the drawing margins are is important. It helps you choose even better middlegame ideas
Only because those exotic endgames are rare in practical play, but they are the most complex, interesting, and hardest part of chess in themselves, and they deserve further study and books written on them, besides the tablebases. I wonder if it will ever be possible to understand the wins in 545 moves...etc, why they win, vs the other 10^1000 possible games from those positions!
I mean... aren't you majoring in comp-sci or math?
By definition they are not the most complex or difficult to understand... the fact that we can solve them on our own / with the help of EGTB means that they're irrefutably some of the easiest positions.
The hardest position is the starting position... but I don't meant this post to be so flippant as that... let's just forget about the opening and pay attention to the middlegame, from the purely maths perspective. These positions are more difficult than pawnless endgames because the game tree complexity is larger.
I mean... aren't you majoring in comp-sci or math?
By definition they are not the most complex or difficult to understand... the fact that we can solve them on our own / with the help of EGTB means that they're irrefutably some of the easiest positions.
The hardest position is the starting position... but I don't meant this post to be so flippant as that... let's just forget about the opening and pay attention to the middlegame, from the purely maths perspective. These positions are more difficult than pawnless endgames because the game tree complexity is larger.
I am a psychology major actually. What I mean is understand why those 545 moves are the best optimal moves possible, other than knowing they were just numbercrunched. Numbercrunching positions doesn't mean you understand the theory, tactic, or the subtle philidor positions occuring through those moves.
Basically, the computer has no clue what it is doing, but can do it anyway.
Sure. Chess is too complex for humans and computer too. That's why humans (and computers) make improvements by finding ways to bypass calculation and use reason (SF nnue > SF)
But still my point remains... the middlegame is closer to the "real" essence of chess...
And that's coming from someone who admired Karpov and Ulf Andersson for years (and still does).
So can LC0 play KNNKP if you do have a video card? Otherwise it's not worth the investment because SF already beats me on full games anyway. I'm looking for something to practise basic endgames against that's nearly but not quite perfect.
It is your choice, it is your money. You do invest whatever you like.
But it is already known Stockfish and Lc0 are two best engines. People who can afford for GPU use both as both have their own weakness and strength.
One of SF8's weaknesses is it can't play KNNKP. SF11 in the guise of Lichess level 8 can't play it either, but it can't play it a lot better than SF8 can't play it, so maybe given a reasonable time control it could.
Does anybody know whether either SF11 or LC0 with GPU can play the endgame given reasonable time controls?
I dont normally do KNN KP as there are Tablebases already. Why waste my time and electricity bill to do it?
However, if you are willing to pay me for $50 for 1 hour for me, I can do that for you. Or else, you can find third party websites that run Lc0 service commercially or rent Amazon cloud yourself.