When will 8 piece tablebase be ready?

[Chessqueen]

The 8 piece tablebase take about 1.5 to 2 petabyte, and nowadays it can be easily done in syzygy format. The 1st 6 piece tablebase was Nalimov tablebase finished around 2005, latest the 7 piece tablebase published was Lomonosov tablebase, in August 2012, and it takes up to 140T. so it is estimated that the 8 piece will take at least 10 years since it is huge. Is somebody working on it, and what will be the Elo gain for any engine using versus NOT using it?

Note: The main question is a top engine using a good Opening and possibly the future 8 piece tablebase versus the same engine not using an opening book and no tablebase what Elo difference are we looking at?

[DrCliche]

It might be "easy" in the sense that the code (more or less) exists, and the storage requirements are within reach of a moderately rich person, but you have to actually generate them first. You'd need a computer with ~50-100TB of RAM and ~10-20PB of storage, plus lots of time and a couple mil for the electricity alone. Good luck!

[hgm]

Pawnless 8-men end-games will virtually never occur, and having EGT for those will bring zero Elo. End-games that do occur frequently enough for mishandling to affect Elo wil have many Pawns. EGT for those can be generated with far fewer resources. And the partial EGT you would need in a given game would be such a small fraction of the total EGT (because most Pawn constellations would no longer be reachable from the game position where the EGT first becomes relevant), that you can easily generated on the fly.

So it seems safe to say that pre-computed 8-men EGT will bring zero Elo to a good engine.

[Chessqueen]

It might be "easy" in the sense that the code (more or less) exists, and the storage requirements are within reach of a moderately rich person, but you have to actually generate them first. You'd need a computer with ~50-100TB of RAM and ~10-20PB of storage, plus lots of time and a couple mil for the electricity alone. Good luck!

In that case has anybody done tests with stockfish engine using a good Opening and 7 piece tablebase versus the same engine not using an opening book and no tablebase and after 1000 games compared the result?

[Chessqueen]

It might be "easy" in the sense that the code (more or less) exists, and the storage requirements are within reach of a moderately rich person, but you have to actually generate them first. You'd need a computer with ~50-100TB of RAM and ~10-20PB of storage, plus lots of time and a couple mil for the electricity alone. Good luck!

In that case has anybody done tests with stockfish engine using a good Opening and 7 piece tablebase versus the same engine not using an opening book and no tablebase and after 1000 games compared the result?

I started a tournament Match of 1000 games both using Stockfish 13 one with an Opening and 7 piece tablebase and the other without Opening nor Tablebase TC 2'1" so far I have NOT noticed any difference they are even after 70 games 35 to 35

[Nordlandia]

I inquired the question on rybka forum few years back. Believe it or not, there is still people dwelling there.

I got the responce that it was up for consideration but was put on the shelf because of the daunting computer resources needed for 8-man project.

https://rybkaforum.net/cgi-bin/rybkafor ... ?tid=26893

[Ajedrecista]

Hello:

I inquired the question on rybka forum few years back. Believe it or not, there is still people dwelling there.

I got the responce that it was up for consideration but was put on the shelf because of the daunting computer resources needed for 8-man project.

https://rybkaforum.net/cgi-bin/rybkafor ... ?tid=26893

I think that thread was an April's Fool Day joke, just looking at the date of the original post and some key words seen later through the thread such as 'fool' and 'hoax'.

Despite not knowing anything about EGTB generation, it seems that the biggest bottleneck would be RAM, let alone a good amount of dollars or whatever currency in bills, as pointed before. Furthermore, the current existing code already supports 8-man EGTB generation or does it need some rewrite?

There is an excellent site about number of unique legal positions that could be of interest here:

NULP (Number of Unique Legal Positions in chess endgames)

Code: Select all

Pieces               Positions
------------------------------
  2                        462
  3                    368,079
  4                125,246,598
  5             25,912,594,054
  6          3,787,154,440,416
  7        423,836,835,667,331
  8     38,176,306,877,748,245

There is even a split of the number of NULP among the 2,520 8-man EGTB:

http://kirill-kryukov.com/chess/nulp/re ... ndgame.txt

The maximum number of NULP is 87,076,702,767,652 (8.71e+13 from KRBNPvKBN) while the minimum is 29,598,163,317 (2.96e+10 from KQQQQQQvK) if I am right. Of course you can select a few balanced EGTB like KRPPvKRPP, KQPPvKQPP, KPPPvKPPP and their promotions to Q and N and plenty of endgames are covered:

Code: Select all

 Endgame            NULP
-------------------------------
KQPPvKQPP     4,139,161,281,300
KRPPvKRPP     4,855,950,632,029
KPPPvKPPP       349,857,828,160

A good exercise would be compute bytes/NULP ratios for each 3, 4, 5, 6 and 7-man Syzygy EGTB and try to search patterns, then estimate 8-man EGTB sizes. I will not do it for each endgame, only for the full sets:

Code: Select all

Total sizes from CPW source (WDL + DTZ).

 Men        Size          NULP        Bytes/NULP
------------------------------------------------
3,4,5     939.0 MiB    2.6038e+9        0.0378
  6       149.2 GiB    3.7872e+12       0.0423
  7        16.7 TiB    4.2384e+14       0.0433
  8         N/A        3.8176e+16        N/A

If a ratio of 0.05 bytes/NULP is assumed, then the whole set of 8-man EGTB would be circa 1.7 PiB in size, way below 10 to 20 PiB mentioned above if my fast math is not wrong. If we keep this ratio, KQPPvKQPP would be circa 192.74 GiB, KRPPvKRPP would be circa 226.12 GiB and KPPPvKPPP would be circa 16.29 GiB. They are only educated guesses.

Regards from Spain.

Ajedrecista.

[Paloma]

When will 8 piece tablebase be ready?

Never?

[MikeB]

I believe they may already exist in this universe, just not in our solar system...

We just need to track down the civilization that invented these flying machines.

[JohnWoe]

Unfortunately I don't think there's other civilizations. At least not inside our observable universe. What's outside nobody knows. The whole universe is fine tuned for life on just one planet. Earth. By Dr. Hugh Ross.
Of course if you could only reap good stuff of 8 egtb. That's would be easy +Elo. But the slowdown on probing won't give much Elo anyway.