How did this happen? SF says +154 and draw.

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Nay Lin Tun
Posts: 700
Joined: Mon Jan 16, 2012 5:34 am

How did this happen? SF says +154 and draw.

Post by Nay Lin Tun » Tue Jul 07, 2020 1:34 am

I have never seen such draw. ( ccc game)




Alayan
Posts: 539
Joined: Tue Nov 19, 2019 7:48 pm
Full name: Alayan Feh

Re: How did this happen? SF says +154 and draw.

Post by Alayan » Tue Jul 07, 2020 2:03 am

Probably Graph History Interaction with the hash table unaware of the 50 move rule.

Milos
Posts: 4064
Joined: Wed Nov 25, 2009 12:47 am

Re: How did this happen? SF says +154 and draw.

Post by Milos » Tue Jul 07, 2020 2:20 am

Alayan wrote:
Tue Jul 07, 2020 2:03 am
Probably Graph History Interaction with the hash table unaware of the 50 move rule.
It's not only that. On move 73.Qd6 is a sixth choice in mpv even with only 5-men. Out of 26 possible moves 5 are clear wins and one is a cursed win. And SF chooses a cursed win. So that's not only a hash miss. It could be an actual bug.

jhellis3
Posts: 488
Joined: Fri Aug 16, 2013 10:36 pm

Re: How did this happen? SF says +154 and draw.

Post by jhellis3 » Tue Jul 07, 2020 2:38 am

There are 8m TBs now?

jhellis3
Posts: 488
Joined: Fri Aug 16, 2013 10:36 pm

Re: How did this happen? SF says +154 and draw.

Post by jhellis3 » Tue Jul 07, 2020 3:02 am

Perhaps I am missing something, but AFAICS (with single thread crystal) the position appears to be winning?

White shuffles a bit to get the queen to e5 threatening the g5 pawn with the black king already at g6, black moves rg8 to hold the pawn, white Qa1 and gg?



zullil
Posts: 6442
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: How did this happen? SF says +154 and draw.

Post by zullil » Tue Jul 07, 2020 5:41 am

Nay Lin Tun wrote:
Tue Jul 07, 2020 1:34 am
I have never seen such draw. ( ccc game)



The position is winning for White, but not within the context of this game, due to three-fold repetition constraints?

Stockfish-dev says White erred with 73. Qd6, which threw away a win with 73. Qd5+.


Uri Blass
Posts: 8975
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: How did this happen? SF says +154 and draw.

Post by Uri Blass » Tue Jul 07, 2020 10:01 am

Alayan wrote:
Tue Jul 07, 2020 2:03 am
Probably Graph History Interaction with the hash table unaware of the 50 move rule.

Maybe it is better to start storing the 50 move rule counter in the hash tables to prevent the GHI problem in these cases when the idea is not to do it in every position but only in positions when all pawns or almost all pawns are blocked.

Joerg Oster
Posts: 816
Joined: Fri Mar 10, 2006 3:29 pm
Location: Germany

Re: How did this happen? SF says +154 and draw.

Post by Joerg Oster » Tue Jul 07, 2020 10:14 am

zullil wrote:
Tue Jul 07, 2020 5:41 am
Nay Lin Tun wrote:
Tue Jul 07, 2020 1:34 am
I have never seen such draw. ( ccc game)



The position is winning for White, but not within the context of this game, due to three-fold repetition constraints?

Stockfish-dev says White erred with 73. Qd6, which threw away a win with 73. Qd5+.
Hmm, a MultiPV=8 search implies there are 7 moves still winning, also Qd6 ...
(1 Thread, 512 GB Hash, syzygy-6)

Code: Select all

info depth 57 seldepth 56 multipv 1 score cp 15401 wdl 1000 0 0 nodes 974738899 nps 1571867 hashfull 1000 tbhits 39469258 time 620115 pv e6d5 b7b6 d5a2 c8c7 d7e6 c7c8 a2a5 b6c6 a5a6 c6c7 a6a7 c7c6 a7c5 c6b7 c5b5 b7c7 b5a5 c7c6 a5a6 c6c7 a6d6 c7b7 e6d7 g8g5 b4b5 g5b5
info depth 57 seldepth 53 multipv 2 score cp 15400 wdl 1000 0 0 nodes 974738899 nps 1571867 hashfull 1000 tbhits 39469258 time 620115 pv e6e2 c8c7 d7e6 c7c4 e6f7 c4c8 e2b5 b7c7 b5c5 c7d7 c5f5 d7d6 f5e6 d6c7 b4b5 g8h8 e6c6 c7b8 c6b6 b8a8 b6a6 a8b8 b5b6 h8h7 f7e6 c8c6 e6d5 h7e7
info depth 57 seldepth 64 multipv 3 score cp 15399 wdl 1000 0 0 nodes 974738899 nps 1571867 hashfull 1000 tbhits 39469258 time 620115 pv e6e4 b7b6 e4d4 b6b7 d4d5 b7b6 d5a2 c8c7 d7e6 c7c8 a2a5 b6c6 a5a6 c6c7 a6a7 c7c6 a7c5 c6b7 c5b5 b7c7 b5a5 c7c6 a5a6 c6c7 a6d6 c7b7 e6d7 g8g5 b4b5 g5b5
info depth 57 seldepth 51 multipv 4 score cp 15399 wdl 1000 0 0 nodes 974738899 nps 1571867 hashfull 1000 tbhits 39469258 time 620115 pv e6e5 b7b6 e5d4 b6b7 d4d5 b7b6 d5a2 c8c7 d7e6 c7c8 a2a5 b6c6 a5a6 c6c7 a6a7 c7c6 a7c5 c6b7 c5b5 b7c7 b5a5 c7c6 a5a6 c6c7 a6d6 c7b7 e6d7 g8g5 b4b5 g5b5
info depth 57 seldepth 49 multipv 5 score cp 15397 wdl 1000 0 0 nodes 974738899 nps 1571867 hashfull 1000 tbhits 39469258 time 620115 pv e6f7 g8h8 f7d5 b7b6 d5e5 h8g8 e5d4 b6b7 d4d5 b7b6 d5a2 c8c7 d7e6 c7c8 a2a5 b6c6 a5a6 c6c7 a6a7 c7c6 a7c5 c6b7 c5b5 b7c7 b5a5 c7c6 a5a6 c6c7 a6d6 c7b7 e6d7 g8g5 b4b5 g5b5
info depth 57 seldepth 84 multipv 6 score cp 15397 wdl 1000 0 0 nodes 974738899 nps 1571867 hashfull 1000 tbhits 39469258 time 620115 pv e6d6 g8h8 d6d5 b7b6 d5e5 h8g8 e5d4 b6b7 d4d5 b7b6 d5a2 c8c7 d7e6 c7c8 a2a5 b6c6 a5a6 c6c7 a6a7 c7c6 a7c5 c6b7 c5b5 b7c7 b5a5 c7c6 a5a6 c6c7 a6d6 c7b7 e6d7 g8g5 b4b5 g5b5
info depth 57 seldepth 49 multipv 7 score cp 15397 wdl 1000 0 0 nodes 974738899 nps 1571867 hashfull 1000 tbhits 39469258 time 620115 pv e6f5 c8c7 d7e6 b7a6 f5f7 c7c8 e6d7 a6b7 f7d5 b7b6 d5a2 c8c7 d7e6 c7c8 a2a5 b6c6 a5a6 c6c7 a6a7 c7c6 a7c5 c6b7 c5b5 b7c7 b5a5 c7c6 a5a6 c6c7 a6d6 c7b7 e6d7 g8g5 b4b5 g5b5
info depth 57 seldepth 28 multipv 8 score cp 0 wdl 32 935 33 nodes 974738899 nps 1571867 hashfull 1000 tbhits 39469258 time 620115 pv e6b3 c8c7 d7e6 c7c6 e6f5 g8e8 b3f7 c6d6 f5f4 b7c7 f7e8 d6d4 f4g5 d4d5 g5g6 d5d6 g6f7
bestmove e6d5 ponder b7b6
However, this probably requires a much deeper search ...
Jörg Oster

zullil
Posts: 6442
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: How did this happen? SF says +154 and draw.

Post by zullil » Tue Jul 07, 2020 10:31 am

OK. I didn't realize the +150 evaluations occurred during a game. Note that the "blue" Stockfish that was kibitzing noticed when Stockfish erred at move 73.

Post Reply