From other forum:
1 Stockfish 6 Man TB 2 1500 41 33 1426 754.0 50.3% 95.1%
2 Stockfish 7 Man TB 1 1500 43 37 1420 753.0 50.2% 94.7%
3 Stockfish 5 Man TB -1 1500 39 44 1417 747.5 49.8% 94.5%
4 Stockfish No TB -2 1500 31 40 1429 745.5 49.7% 95.3%
!?
7 piece tablebase "benefit"
Moderators: hgm, Rebel, chrisw
-
- Posts: 12566
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: 7 piece tablebase "benefit"
I guess that the disk subsystem is very important.
I only have a partial set of the 7 man files, but they live on super fast gumstick SSD.
If you have ultra-fast SSD for 6 man, and ordinary spinning platters for 7 man, then the result is not surprising at all.
I only have a partial set of the 7 man files, but they live on super fast gumstick SSD.
If you have ultra-fast SSD for 6 man, and ordinary spinning platters for 7 man, then the result is not surprising at all.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
- Posts: 710
- Joined: Sat Dec 06, 2014 1:53 pm
Re: 7 piece tablebase "benefit"
I have partial subset: 7-man WDL-only on 500 Gb Samsung 870 EVO + full 5/6-man (on other NVME SSD)
Total about 700 Gb
7-man subset ** I took from this rating, first 70 files or so
I run 4CPU SF-dev (Abrok latest AVX2) with this set (5 + 6 + partial 7-man WDL) and no-TB.
1000 games on 1' + 1" and 500 games on 3' + 2"
noob_3moves.epd (fishtest), RepeatStartposition=YES
0 ELO difference in both runs
** Excel file created from this topic, rating by Zenmastur
http://talkchess.com/forum3/viewtopic.php?f=2&t=69309
Total about 700 Gb
7-man subset ** I took from this rating, first 70 files or so
I run 4CPU SF-dev (Abrok latest AVX2) with this set (5 + 6 + partial 7-man WDL) and no-TB.
1000 games on 1' + 1" and 500 games on 3' + 2"
noob_3moves.epd (fishtest), RepeatStartposition=YES
0 ELO difference in both runs
** Excel file created from this topic, rating by Zenmastur
http://talkchess.com/forum3/viewtopic.php?f=2&t=69309
-
- Posts: 4889
- Joined: Thu Mar 09, 2006 6:34 am
- Location: Pen Argyl, Pennsylvania
Re: 7 piece tablebase "benefit"
The real benefit of using 7 men TB ( for me, ymmv) is knowing whether the position is a cursed win or not. Cursed wins are real wins under ICCF rules. For that you may need to have your engine and/or GUI configured differently.
In this example, only Re5 wins (under ICCF rules).
[d]6r1/8/8/6N1/2k2r2/6R1/4R3/3K4 w - - 0 1
https://syzygy-tables.info/?fen=6r1/8/8 ... _w_-_-_0_1
So if that is important to you, you would want to see this output on the above position:
Code: Select all
36 +100.00 451.7M 0:17.05 Re5 Rgf8
35 +100.00 394.8M 0:15.04 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb1+ Kf2 Rb5 Re8 Rf5+ Ke1 Kb5 Ra2 Kc6 Nf2 Rg3 Ke2 Kc7 Re6 Kb8 Re4 Kc7 Rc4+ Kb6 Rca4 Re5+ Kf1 Rf5 Ra7 Rgf3 Re7 Rg3 Rb2+ Kc6 Rd2 Rf8 Rde2 Rg6 Ra2 Kb6 Rc2
34 +100.00 136.1M 0:05.11 Re5
33 +100.00 91.3M 0:03.44 Re5
32 +100.00 63.7M 0:02.40 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Kf1 Rh2 Kg1 Rc2 Ra4+ Kb3 Rf4 Kb2 Ref5 Ka1 Nf2 Rb1+ Kg2 Rbc1 Kg3 Rg1+ Kh4 Rc8 Rh5 Rc2 Kh3 Rb2 Kh2 Rf1 Kg2 Re1 Rf7 Ra2 Rc5 Rb1
31 +100.00 32.2M 0:01.22 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb3 Ra4+ Rb4 Raa5 Rb2 Kf1 Kb3 Reb5+ Rb4 Rf5 Ra4 Rab5+ Ka2 Rbe5 Rc4 Nf2 Rc1+ Kg2 Ka1 Rd5 Re1 Rd4 Rc2 Rf3 Rec1 Ra4+ Ra2 Raf4 Re1 Rd4 Rc1 Rdd3 Rb1 Rf4 Re1 Kh3 Rf1
30 +100.00 23.4M 0:00.90 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb3 Ra4+ Rb4 Raa5 Rb2 Kf1 Kb3 Reb5+ Rb4 Rf5 Ra4 Rab5+ Ka2 Rbe5 Rc4 Nf2 Rc1+ Kg2 Ka1 Rd5 Ra2 Rf3 Re1 Rdd3 Rc2 Rf7 Rec1 Rdf3 Rb2 R7f4 Ra2 Rd3 Rac2 Kf3 Rf1 Kg3
29 +100.00 12.9M 0:00.49 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rc3 Reb1 Re3 Rc1 Rf5 Rbc2 Ref3 Re1 Rb5 Rec1 Rf4 Ra2 Rd5
28 +100.00 10.9M 0:00.42 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rf7 Ra2 Rc3 Rb2 Rf8 Reb1 Rf4 Re1 Kg3 Rg1+ Kh3 Rf1 Ra3+ Ra2 Raf3
27 +100.00 9.70M 0:00.37 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rf7 Ra2 Rcc7 Rb1 Rc3 Re1 Rf4 Rb2 Kg3 Rf1 Ra3+ Ra2 Re3 Rb2 Kg2
26 +100.00 8.35M 0:00.31 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rf7 Ra2 Rcc7 Kb1 Rc4 Rb2 Rc3 Ka1 Kg3 Rf1 Rcf3 Ra2 Kh3 Rb2 Kg2
25 +100.00 8.03M 0:00.30 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rf7 Ra2 Rcc7 Kb1 Rc4 Rb2 Rc3 Ka1 Kg3 Rf1 Rcf3 Ra2
24 +100.00 7.93M 0:00.30 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rf7 Ra2 Rcc7 Kb1 Rc4 Rb2 Rc3 Ka1 Kg3 Rf1
23 +100.00 7.83M 0:00.29 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rf7 Ra2 Rcc7 Kb1 Rc4 Rb2 Rc3 Ka1 Kg3 Rf1 Rcf3 Ra2 Kh3 Rb2 Kg2
22 +100.00 7.75M 0:00.29 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rf7 Ra2 Rcc7 Kb1 Rc4 Rb2 Rc3 Ka1 Kg3 Rf1 Rcf3
21 +100.00 7.66M 0:00.29 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rg2 Ra4+ Kb3 Rf4 Ka2 Ra5+ Kb1 Kf1 Ra2 Rd5 Ka1 Nf2 Rb1+ Kg2 Rab2 Rc5 Re1 Rf7 Ra2 Rcc7 Kb1 Rc4 Rb2 Rc3 Ka1 Rd3
20 +100.00 661190 0:00.02 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Rf4 Kb2 Rf7 Ka1 Nf2 Ra2 Re3 Rb1+ Kg2 Rc2 Ra3+ Ra2 Rc3 Re1 Rcf3 Rc1 R7f4 Rb2 Rf5 Ra2
19 +100.00 607682 0:00.02 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Ra7 Kc3 Rc7+ Kb2 Rf7 Ka1 Nf2 Ra2 Re3 Rb1+ Kg2 Rc2 Ra3+ Ra2 Rc3 Re1
18 +100.00 565597 0:00.02 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Rf4 Kb2 Rf7 Ka1 Nf2 Ra2 Re3 Rb1+ Kg2 Rc2 Ra3+ Ra2 Rc3 Re1
17 +100.00 409957 0:00.02 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Ra7 Kc3 Rc7+ Kb2 Rf7 Ka1 Nf2 Rb1+ Kg2 Rd2 Ra8+ Ra2 Rc8 Rd2
16 +100.00 377218 0:00.01 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Ra7 Kc3 Rc7+ Kb2 Rf7 Ka1 Nf2 Rb1+ Kg2 Rd2 Ra8+ Ra2 Rc8 Rd2
15 +100.00 313658 0:00.01 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Rf4 Kb2 Rf7 Ka1 Nf2 Rb1+ Kg2 Rd2 Ra8+ Ra2 Rc8 Rd2
14 +100.00 298621 0:00.01 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Rf4 Kb2 Nf2 Ka1 Re3 Rb1+ Kg2 Rbc1 Ra3+ Ra2 Rd3 Rac2 Kg3 Rg1+ Kh3 Rf1 Ra3+ Ra2
13 +100.00 196404 0:00.01 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Rf4 Kb2 Rf7 Ka1 Nf2 Rb1+ Kg2 Rd2 Ra8+ Ra2 Rc8 Re1
12 +100.00 173273 0:00.01 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Rf4 Kb2 Ng1 Rc1+ Kf2 Rcc5 Nf3 Rc2+ Kf1 Ka1 Ra4+ Ra2
11 +100.00 82549 0:00.00 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Rf4 Kb2 Nf2
10 +100.00 67183 0:00.00 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rc2 Ra4+ Kb3 Rf4 Kb2 Ng1 Rc1+ Kf2
9 +100.00 56674 0:00.00 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rb2 Ra4+ Rb4 Ra5 Rb3
8 +100.00 21959 0:00.00 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3 Rb5 Re8 Rg2 Kf1 Rh2
7 +100.00 13170 0:00.00 Re5 Rd4+ Ke1 Rb8 Ra3 Rg4 Nh3
6 +100.00 7669 0:00.00 Re5 Rgf8 Kd2 Rd4+ Ke1 Rb8 Ra3
5 +100.00 4376 0:00.00 Re5 Rgf8 Kd2 Rd4+ Ke1 Ra8
4 +100.00 2574 0:00.00 Re5 Rgf8 Kd2 Rd4+ Ke1
3 +100.00 1550 0:00.00 Re5 Rgf8 Kd2
2 +100.00 759 0:00.00 Re5 Rgf8
1 +100.00 245 0:00.00 Re5
-
- Posts: 5627
- Joined: Wed Sep 05, 2018 2:16 am
- Location: Moving
- Full name: Jorge Picado
Re: 7 piece tablebase "benefit"
So today's super engine like Stockfish Komodo dragon and LCZero, really do NOT need any EGTB when facing lower rated engine, therefore, I believe that those three engines in particular should NOT use EGTB when being rated and when they play lower rated engines below 3300yurikvelo wrote: ↑Tue Mar 30, 2021 9:52 am I have partial subset: 7-man WDL-only on 500 Gb Samsung 870 EVO + full 5/6-man (on other NVME SSD)
Total about 700 Gb
7-man subset ** I took from this rating, first 70 files or so
I run 4CPU SF-dev (Abrok latest AVX2) with this set (5 + 6 + partial 7-man WDL) and no-TB.
1000 games on 1' + 1" and 500 games on 3' + 2"
noob_3moves.epd (fishtest), RepeatStartposition=YES
0 ELO difference in both runs
** Excel file created from this topic, rating by Zenmastur
http://talkchess.com/forum3/viewtopic.php?f=2&t=69309
Who is 17 years old GM Gukesh 2nd at the Candidate in Toronto?
https://indianexpress.com/article/sport ... t-9281394/
https://indianexpress.com/article/sport ... t-9281394/
-
- Posts: 1470
- Joined: Mon Apr 23, 2018 7:54 am
Re: 7 piece tablebase "benefit"
The question is: can SF without 7-man TB convert this position (or positions with shorter but still long lines) to a win? (Some modification of the 50-move rule code may be necessary.)MikeB wrote: ↑Tue Mar 30, 2021 6:19 pm The real benefit of using 7 men TB ( for me, ymmv) is knowing whether the position is a cursed win or not. Cursed wins are real wins under ICCF rules. For that you may need to have your engine and/or GUI configured differently.
In this example, only Re5 wins (under ICCF rules).
[d]6r1/8/8/6N1/2k2r2/6R1/4R3/3K4 w - - 0 1
https://syzygy-tables.info/?fen=6r1/8/8 ... _w_-_-_0_1
-
- Posts: 2821
- Joined: Fri Sep 25, 2015 9:38 pm
- Location: Sortland, Norway
Re: 7 piece tablebase "benefit"
7-piece even partially do clearly make some difference compared to 6-pieces.
6 men + partial 7 piece, like most important tables do certainly provide more help.
6 men + partial 7 piece, like most important tables do certainly provide more help.
-
- Posts: 12566
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: 7 piece tablebase "benefit"
I think it probably comes down to this:
In a very sparse board, most strong engines will win in a won position.
Further, most games are decided before the board is extremely sparse.
There are, of course, positions where a tablebase will mean the difference between winning and not winning.
For me, I use them for analysis.
Given enough time you get really good results.
They also tend to make for prettier endgames (especially with DTM tablebase files)
There are many things that do not lead to more Elo in fast games.
Removal of under-promotion, especially to bishop, is one example.
Perhaps one game in 100,000 might be affected.
But I want the right answer,darn it, even if it costs me 19 Elo.
I expect that tablebase files matter more on really high end systems.
For TCEC class machines, you get lots of tablebase hits from the opening position, given a few minutes of thinking.
From ten full moves or more, you get enormous numbers of tablebase hits.
From late midgame forward, the engines spend most of their time "talking to God"
In a very sparse board, most strong engines will win in a won position.
Further, most games are decided before the board is extremely sparse.
There are, of course, positions where a tablebase will mean the difference between winning and not winning.
For me, I use them for analysis.
Given enough time you get really good results.
They also tend to make for prettier endgames (especially with DTM tablebase files)
There are many things that do not lead to more Elo in fast games.
Removal of under-promotion, especially to bishop, is one example.
Perhaps one game in 100,000 might be affected.
But I want the right answer,darn it, even if it costs me 19 Elo.
I expect that tablebase files matter more on really high end systems.
For TCEC class machines, you get lots of tablebase hits from the opening position, given a few minutes of thinking.
From ten full moves or more, you get enormous numbers of tablebase hits.
From late midgame forward, the engines spend most of their time "talking to God"
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
- Posts: 4889
- Joined: Thu Mar 09, 2006 6:34 am
- Location: Pen Argyl, Pennsylvania
Re: 7 piece tablebase "benefit"
Stockfish has a UCI option to turn off the 50 move rule for 7 men or less , which works on the command line, but whether it works in a GUI is dependant on the specific GUI being used.Some GUIs enforce the 50 move rule regardless if the engine says to ignore it.jp wrote: ↑Thu Apr 01, 2021 9:45 amThe question is: can SF without 7-man TB convert this position (or positions with shorter but still long lines) to a win? (Some modification of the 50-move rule code may be necessary.)MikeB wrote: ↑Tue Mar 30, 2021 6:19 pm The real benefit of using 7 men TB ( for me, ymmv) is knowing whether the position is a cursed win or not. Cursed wins are real wins under ICCF rules. For that you may need to have your engine and/or GUI configured differently.
In this example, only Re5 wins (under ICCF rules).
[d]6r1/8/8/6N1/2k2r2/6R1/4R3/3K4 w - - 0 1
https://syzygy-tables.info/?fen=6r1/8/8 ... _w_-_-_0_1
-
- Posts: 1470
- Joined: Mon Apr 23, 2018 7:54 am
Re: 7 piece tablebase "benefit"
IIRC (from previous discussion) that SF option does not do what we want.MikeB wrote: ↑Fri Apr 02, 2021 5:18 am Stockfish has a UCI option to turn off the 50 move rule for 7 men or less , which works on the command line, but whether it works in a GUI is dependent on the specific GUI being used. Some GUIs enforce the 50 move rule regardless if the engine says to ignore it.
Update: Yes, IRC. I found relevant posts...
We want an option to switch off the 5-move rule, with or without any TBs, whenever we want.Joerg Oster wrote: ↑Mon Mar 16, 2020 10:59 amYes, in regular Stockfish this option only affects the handling of TBs.Zenmastur wrote: ↑Mon Mar 16, 2020 12:20 amThanks Joerg. Do you know when using “regular” Stockfish if you turn off the Syzygy50Moverule does the main program still enforce the 50-move-rule. It doesn't seem make sense that it would. I mean, what good would it do to switch it off for TB's and still have the program enforce the rule?Joerg Oster wrote: ↑Sun Mar 15, 2020 10:35 pm You might give Moonfish a try. https://github.com/joergoster/Moonfish
If you set "Syzygy50MoveRule" to false, this will also switch it off for the search.
If it does, it seems like this is a bug and needs to be changed.
The search still enforces the 50-move rule.
I don't think this is a bug but intentional.