20

An earlier question here asked in 2012 about the status of 7-piece tablebases. Today complete 7-piece tables are widely available.

So today, what is the status of 8-piece endgame tablebases?

  • How complete is the solution?
  • Where are they available? (especially for free)
  • What interesting results have emerged so far?

UPDATE:

Here in Chessbase is an interview dated Nov-2022 with Marc Bourzutschky which indicates the status is perhaps halfway complete.

Laska
  • 11,846
  • 4
  • 44
  • 76
  • 2
    This question currently being discussed on Meta. – SecretAgentMan Jul 06 '23 at 12:07
  • Just curious about Meta angle on the same thing. What is chess meta about that chess can't address. Is it a level of some kind? – dbdb Nov 16 '23 at 00:40
  • 1
    I think it was “piece” vs “man” iirc. I think the former is actually preferable, so I edited the question – Laska Nov 21 '23 at 17:47
  • Well, while all chess mobile unit instance on board might be pointed at as pieces of material they are made of, the poor pawns are terminologically excluded from being pieces. So it might also be a factor in the meta chess word discussion point of view. I also like piece. Maybe lowercase. and uppercase restricted to non pawn? – dbdb Nov 22 '23 at 20:40
  • 1
    “Piece” has two meanings, so in principle I would prefer “unit” but the intended meaning is clear here – Laska Nov 23 '23 at 01:26

2 Answers2

17

Interesting points:

  1. The longest sequence so far is 584 moves to win.
    • In the 7-piece tablebase, the longest sequence is 549 moves, so it is surprising that the 8-piece tablebase is not too much longer.
    • This sequence involves white having 2 dark-squared bishops, so it is not a position that is likely to occur in real games.
    • They are confident that they've examined the longest sequences in pawnless endgames, but they may still find longer sequences in endgames involving pawns (such as the 584 moves to win position).
  2. The incomplete database is too large to feasibly share at the moment.
  3. A new, interesting full-point zugzwang position has been discovered (FEN: 3R4/8/2q5/8/8/2k1r3/2r5/1R1K1Q2).
    • If it is White to move, Black can force a win, if it is Black to move, White can force a win.
    • Marc is interested in such positions which do not involve pawns, knights, or promotions. No such position was found in smaller tablebases, but there is one in the 7-piece tablebase that comes close (it involves two same-colored bishops and thus requires an under-promotion to reach).
    • Thusfar, the only other known position matching his criteria was a ten-piece study composed by Javier Rodriguez Ibran (FEN: 8/8/8/8/8/B6b/1r4R1/1rkqQKR1).
Nelson O
  • 705
  • 3
  • 6
  • 2
    "...so it is surprising that the 8-piece tablebase is not too much longer..." There exists an informal paper (needs citation) expecting exactly this. Argument in short: forced exchanges. Note that in the given position no sensible exchanges exist. (Sorry to be cryptic, I'm quoting from memory.) – Hauke Reddmann Jul 01 '23 at 11:28
12

We are still far away from the solution. Estimates for the size of the completed tablebase are around 10 Petabytes. Compare that to the 140 Terabyte hard drive (later trimmed down to 18TB in the syzygy table) and we are still not anywhere near achieving it.

I don’t think that it couldn’t be done this year however, my gut tells me that it’s more due to a lack of trying. Supercomputers are rarely used for chess anymore and someone would have to convince probably a billionaire to “waste” millions of dollars worth of super computing time to achieve it. Once the tablebase is finished, how does one profit off it. Will they even get their money back?

The most interesting find to me at the moment is this world record 8 piece 584 move forced checkmate. It was found by Marc Bourzutschky last year in as far as I can tell the most recent attempt at the 8 piece tablebase. It contains 2 white dark square bishops.

R7/8/8/8/7q/2K1B2p/7P/2Bk4 w - - 0 1
Rewan Demontay
  • 17,514
  • 4
  • 67
  • 113
Matthew
  • 329
  • 7
  • 2
    Is millions of dollars of supercomputing really needed, or can the project be done in a distributed fashion like they do for Mersenne primes? – qwr Jul 01 '23 at 15:06
  • 1
    Mersenne primes have been found with as little as 150 laptops, for this it would take a number in the thousands. I imagine that a famous (rich) person would have to sponsor it, leaving you with the same problem as before. – Matthew Jul 03 '23 at 07:31
  • 2
    @qwr the main problem is that you'd probably need 64 TB of RAM to run the standard algorithm. Every decrease of memory compared to that you'll have to pay with heavy runtime increases. And currently 64 TB of RAM is not very widely available to put it mildly. – koedem Jul 07 '23 at 21:42