Stockfish: Major blunder (bug?) in CCC11 semis game 33
FEN: 8/8/6p1/6Pp/4R2P/2K4k/8/3r4 w - - 91 111
Here SF somehow missed Kb4 is losing and blundered while it has partial 7men. Is this a bug?
adentong
All 38 comments
Here is the full game score
Rocky640
on 10 Dec 2019
Version 081219 64 BMI2 is playing with following settings.
Contempt 16
Hash 16384
SyzygyPath /data/tablebases
Threads 90
Rocky640
on 10 Dec 2019
Looks like the GHI issue again. This is a tough one to crack.
MichaelB7
on 10 Dec 2019
It would be useful if we could have a reproducer for this.
As we know GHI would be hard to address, but it would be nice if we could make sure (as opposed to very likely) that it is the issue.
vondele
on 10 Dec 2019
Occyroexanthub wrote the following in CCC chat:
"My latest SF dev with 3-6 full TBs and about half 7 men, had to go to depth 218 (after only 2 seconds though) to avoid 111.Kb4. After 111.Kb4, Rc1 SF has "huge" problems(as it needs to reach depth 153) with TBs to find it's losing.
Analysis by Stockfish 091219 64 POPCNT:
112.Re4-d4 Kh3-g3 113.Kb4-a4 Rc1-a1+ 114.Ka4-b5 Ra1-h1 115.Rd4-a4 The position is equal: = (0.00) Depth: 152/8 00:00:03 472MN, tb=3359
112.Re4-d4 Kh3-g3 113.Kb4-a4 Rc1-a1+ 114.Ka4-b5 Ra1-h1 115.Rd4-a4 The position is equal: = (-0.09 --) Depth: 153/8 00:00:9 1392MN, tb=4599
As of now in 5 runs I've made with TBs (with SF latest of 09Dec2019) in 3 it preferred from the beginning to end 111.Kc4 or Kc2 that draws. In 2 it preferred 111.Kb4 that loses and avoided it after an insane amount of depth(>200 plies) but in 2 seconds.
When I played the blunder 111.Kb4, it always was slow(in depth terms not in time) to realize its losing.
And when I used no TBs it was super fast (from beginning) to avoid Kb4 blunder and also see after playing it that it loses. So maybe this is a TB issue."
crossbr
on 11 Dec 2019
You have the load the game history and with analysis on so that the large hash gets filled , work up to the move in question slowly - that is the only way to come close to duplicating a GHI issue. Hash tables do not have game status - so the position in question could have been scored a draw because of the 50 move rule and stored in TT table - next lookup , 50 move rule does not apply, but the position is still scored a draw in the TT. One way to avoid this is to not store any 50 move rule draws in TT. Not sure if it would cause a regression or not.
MichaelB7
on 11 Dec 2019
For what it is worth, I downloaded the game pgn while at work today, and also downloaded the latest SFdev. My work computer has no TB. When I analyzed the position at move 111, white to move, with hash cleared, and the latest SFdev, but no TB, Stockfish abandoned Kb4 at d=8 (i.e. almost instantly), and never came back to Kb4.
crossbr
on 11 Dec 2019
@crossbr Bryan, that is a sign of the GHI issue as being the culprit.
MichaelB7
on 11 Dec 2019
I'm sure this is a dumb question, but is it too costly (elo-wise) to include each position's 50 move status and repetition status (i.e. 0, 1, or 2) in the hash table, so that the identity of the position hashed is defined also by these two aspects of game history, so that two different TT entries could have two distinct values, because of different history data, and yet both be of the same arrangement of pieces on the board?
crossbr
on 11 Dec 2019
I analyze the game as I described above -
dep score nodes time (not shown: tbhits knps seldep)
111 +4.06! 6.87G 1:42.46 Rc1!
111 +3.05! 6.46G 1:34.28 Rc1!
111 +2.26! 6.14G 1:28.00 Rc1!
111 +1.65! 5.65G 1:19.57 Rc1!
111 +1.17! 5.10G 1:12.00 Rc1!
111 +0.81! 4.34G 1:01.45 Rc1!
111 +0.54! 3.66G 0:52.41 Rc1!
111 +0.34! 2.92G 0:42.63 Rc1!
111 +0.19! 1.98G 0:29.92 Rc1!
111 +0.09! 1.30G 0:20.56 Rc1!
110 0.00 253.8M 0:04.23 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
109 0.00 243.6M 0:04.07 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
108 0.00 217.2M 0:03.64 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
107 0.00 209.2M 0:03.52 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
106 0.00 201.0M 0:03.38 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
105 0.00 193.0M 0:03.24 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
104 0.00 149.0M 0:02.46 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
103 0.00 139.5M 0:02.32 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
102 0.00 137.9M 0:02.29 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
101 0.00 122.0M 0:02.04 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
100 0.00 114.1M 0:01.91 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
99 0.00 113.4M 0:01.90 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
98 0.00 89.7M 0:01.53 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
97 0.00 72.5M 0:01.25 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
96 0.00 66.7M 0:01.16 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
95 0.00 63.1M 0:01.10 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
94 0.00 56.5M 0:00.99 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
93 0.00 53.1M 0:00.93 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
92 0.00 49.6M 0:00.88 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
91 0.00 46.2M 0:00.82 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
90 0.00 43.3M 0:00.77 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
89 0.00 39.3M 0:00.71 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
88 0.00 34.4M 0:00.63 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
87 0.00 30.8M 0:00.57 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
86 0.00 28.2M 0:00.53 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
85 0.00 26.2M 0:00.49 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
84 0.00 23.9M 0:00.45 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
83 0.00 22.5M 0:00.43 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
82 0.00 21.2M 0:00.40 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
81 0.00 19.5M 0:00.38 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
80 0.00 18.1M 0:00.35 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
79 0.00 16.2M 0:00.32 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
78 0.00 15.2M 0:00.30 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
77 0.00 14.3M 0:00.28 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
76 0.00 12.7M 0:00.26 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
75 0.00 11.8M 0:00.24 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
74 0.00 10.8M 0:00.22 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
73 0.00 10.1M 0:00.21 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
72 0.00 9.38M 0:00.19 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
71 0.00 8.82M 0:00.18 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
70 0.00 8.25M 0:00.17 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
69 0.00 7.64M 0:00.16 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
68 0.00 7.19M 0:00.15 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
67 0.00 6.80M 0:00.14 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
66 0.00 6.29M 0:00.13 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
65 0.00 5.94M 0:00.13 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
64 0.00 5.55M 0:00.12 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
63 0.00 5.22M 0:00.11 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
62 0.00 4.90M 0:00.11 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
61 0.00 4.64M 0:00.10 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
60 0.00 4.37M 0:00.10 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
59 0.00 4.10M 0:00.09 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
58 0.00 3.81M 0:00.08 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
57 0.00 3.59M 0:00.08 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
56 0.00 3.34M 0:00.08 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
55 0.00 3.11M 0:00.07 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
54 0.00 2.90M 0:00.07 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
53 0.00 2.69M 0:00.06 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
52 0.00 2.51M 0:00.06 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
51 0.00 2.35M 0:00.06 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
50 0.00 2.20M 0:00.05 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
49 0.00 2.04M 0:00.05 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
48 0.00 1.89M 0:00.05 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
47 0.00 1.78M 0:00.04 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
46 0.00 1.66M 0:00.04 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
45 0.00 1.56M 0:00.04 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
44 0.00 1.45M 0:00.04 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
43 0.00 1.34M 0:00.04 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
42 0.00 1.25M 0:00.03 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
41 0.00 1.18M 0:00.03 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
40 0.00 1.08M 0:00.03 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
39 0.00 1.00M 0:00.03 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
38 0.00 888134 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
37 0.00 811992 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
36 0.00 737040 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
35 0.00 675790 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
34 0.00 615716 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
33 0.00 575251 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
32 0.00 534891 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
31 0.00 501166 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
30 0.00 464381 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
29 0.00 431562 0:00.02 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
28 0.00 388672 0:00.01 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
27 0.00 361538 0:00.01 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
26 0.00 335944 0:00.01 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
25 0.00 202930 0:00.01 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
24 0.00 168635 0:00.01 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
23 0.00 150066 0:00.01 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
22 0.00 131706 0:00.01 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
21 0.00 107600 0:00.01 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
20 0.00 93506 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
19 0.00 79311 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
18 0.00 70559 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
17 0.00 64076 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
16 0.00 55824 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
15 0.00 47028 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
14 0.00 41975 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
13 0.00 37508 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
12 0.00 32319 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
11 0.00 27287 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
10 0.00 22198 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
9 0.00 17616 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
8 0.00 15020 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
7 0.00 10924 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
6 0.00 8071 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4 Rxg6
5 0.00 6317 0:00.00 Rg1 Kc3 Rg4 Re6 Kxh4
4 0.00 4676 0:00.00 Rg1 Kc3 Rg4 Re6
3 +0.33 2822 0:00.00 Rg1 Kc3 Rg4
2 +0.33 1314 0:00.00 Rg1 Kc3
1 0.00 184 0:00.00 Rg1
A few comments, SF played Kb4 at depth 111 in 3.3 seconds . in my analysis above with letting the hash table populate first , SF finds Kb4 is bad a depth 111 in just over 20 seconds. Prior to Kb4, it thought Kb4 was draw , in total, 6 moves draw. From a clean start, SF abandons Kb4 at depth 8, similar to Brian's analysis above. At this point in the game, both sides were making moves in about 3.5 seconds, but fast enough to see 240M nps or 790M nodes for move 111 ( impressive hardware). Draw scores from TT tables based on 50 moves cannot be relied upon. The game was 9 plies from being drawn, ( note the 50 move draw count at 91) , the move 114. ...Rxh4 was exactly 99 plies from the previous capture at 65. Rxa2. There is no question that this 'buggy' move was the result of the GHI issue - storing a TTvalue with draw score based on the 50 move rule that is looked up when the game state is different from how it was stored and can no longer be relied upon. One solution to try is to stop storing draw scores in the tt table that stem from the 50 move rule count ( or make sure they are overwritten relatively quickly) . They are simply not reliable , especially when the 50 move rule count gets close to 100 plies.
MichaelB7
on 11 Dec 2019
For all positions , yes too costly , too many entries , but maybe for only
draw scores where the 50 move ply counter is > 90? might be worth an
effort.
On Tue, Dec 10, 2019 at 10:30 PM Bryan notifications@github.com wrote:
I'm sure this is a dumb question, but is it too costly (elo-wise) it to
include each position's 50 move status and repetition status (i.e. 0, 1, or
2) in the hash table, so that the identity of the position hashed is
defined also by these two aspects of game history, so that two different TT
entries could have two distinct values, because of different history data,
and yet both be of the same arrangement of pieces on the board?—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/official-stockfish/Stockfish/issues/2451?email_source=notifications&email_token=ADINNUZ5STWNBUGQNEO5FZ3QYBNFNA5CNFSM4JZC5PH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGRYFNA#issuecomment-564363956,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ADINNUYXUBIXMQAW5AYE6D3QYBNFNANCNFSM4JZC5PHQ
.
MichaelB7
on 11 Dec 2019
@crossbr in general this is too costly. I've yesterday tried something to use at least a few 'classes' of 50mr, spaced far apart (in 50mr count terms) initially, but closer near the 100 plies. This is likely to lose Elo: http://tests.stockfishchess.org/tests/view/5df017e29adce9e7e53b9b6a but maybe the implementation can be improved (several variants of this were tried in the past by others and myself).
vondele
on 11 Dec 2019
@MichaelB7 this was an attempt to refresh TT entries on higher 50mr counts:
http://tests.stockfishchess.org/tests/view/5d93c76a0ebc594fb88ee80d
and a similar idea trying to skip TT usage from time to time:
http://tests.stockfishchess.org/tests/view/5d5e32b10ebc5925cf1130eb
Maybe some ideas along these lines can be tried a bit more extensively.
vondele
on 11 Dec 2019
A theoretically better solution instead of trying to store 50mr exact counters is to store bounds.
e.g., we search position A with 50mr counter = N and find a draw. We then store this in hash, and for any 50mr counter >= N we return a draw too (higher 50mr counter can only make it more drawish). But if 50mr counter < N, then we don't know if the draw value is valid, and we do another search. If for M < N, we find out that it's still a draw, we can update the bound to M. If instead we find that it's winning with counter = M, then we can store this as a second bound and assume that it's also winning with counter <= M (lower 50mr counter can never make it any less winning). If the position is then searched with 50mr counter between M and N, we'd have to do another search and update one of the bounds.
Of course, SF eval often doesn't tell us "this is winning" and "this is drawn", so a difficulty with this approach arise if we'd find a lot of steps, i.e. search eval going slowly down with higher 50mr counter. We don't really want to store more than two values, so if .eg. we get +2 for count A, +1 for count B > A, and 0 for count C > B, there has to be some implementation decision to know what to do, and it will be imperfect.
For endgame positions like the one behind this issue, though, with TB usage, a draw bound and a win bound only would work fine.
This still involves significant overhead, but it doesn't destroy TT efficiency anywhere as much as using exact 50mr would.
In theory, some positions would suffer from extra overhead because we hit the position with ascending/descending 50mr counters. But in practice, most positions won't be anywhere as unlucky.
A difficulty is that GHI is very minor at fishtest conditions while the overhead is noticeable. In CCC conditions, GHI hurts more (though outright blunders that go into the PV are rare, the search tree has many), while the overhead would be less significant.
As it mostly affects endgames, enabling it only if piece count on board <= some X might be the best compromise between overhead and accuracy. All the major GHI losses I remember had not much pieces left on board.
Alayan-stk-2
on 11 Dec 2019
Very astute observations, just to add, the major GHI losses ,very few pieces but always more than 7, very high 50 move counts, 90 ply or more and many very deep searches, 100 ply is reached very quickly. In the given instance , the last piece capture was 91 ply ago so I believe the most common occurrences are when there are 10 or fewer movable pieces. We have also seen it happen with more pieces when there are several pawns that are locked up each other where the search depth also rises very quickly.
MichaelB7
on 11 Dec 2019
@MichaelB7 this was an attempt to refresh TT entries on higher 50mr counts:
http://tests.stockfishchess.org/tests/view/5d93c76a0ebc594fb88ee80d
and a similar idea trying to skip TT usage from time to time:
http://tests.stockfishchess.org/tests/view/5d5e32b10ebc5925cf1130ebMaybe some ideas along these lines can be tried a bit more extensively.
@vondele the one test did quite well, 1.63 Elo gain , 91% LOS. To test this on fishtest , it may require us to turn off adjudication and to allow sufficient time to get very deep searches. The time control of 10 sec plus .1 second increment is hardly the appropriate time control to fix this.
MichaelB7
on 11 Dec 2019
Are we sure this is not a nasty side-effect of using incomplete 7-syzygy set?
I was not able to reproduce this with complete 6-syzygy bases.
joergoster
on 11 Dec 2019
How quick can we clear hash tables if 90 threads
and Hash 16384
If we can't rely on the TT tables in some situations, maybe it would be better to clear hash table after every move when high 50 moves count or some other extreme condition. But this would be quite hard to prove that it actually help, and quite hard to avoid cases where it would not help.
Rocky640
on 11 Dec 2019
@joergoster @crossbr reproduced this without TB. I'd still like to see a precise sequence of UCI commands (non-threaded) leading to this.
@Rocky640 cleaning the hash is relatively fast, but the penalty would be huge, I suspect. We do have the TTgeneration available, so we could take more precise action if we had an idea what...
vondele
on 11 Dec 2019
Are we sure this is not a nasty side-effect of using incomplete 7-syzygy set?
I was not able to reproduce this with complete 6-syzygy bases.
I was able to reproduce with complete 7 Egtb, you load the game , start with move 1 and move slowly forward the game with analysis on until the move in question filling the hash , at that juncture , Kb4 was rated draw for about 20 seconds on my machine - but with clear hash it’s very quick to see Kb4 loses.
MichaelB7
on 11 Dec 2019
"and a similar idea trying to skip TT usage from time to time:"
This test didn't go so bad actually. Maybe this could be combined with @MichaelB7's observations. Ie. increase probability for higher 50 move rule values, or only enable random discards when 50 move rule count is > threshold.
Clearing the hash is imo not an option because it may take seconds when the hash sizes reach tens of gigs. It would completely change the asymptotic time complexity of the search. Alternatively entries from the previous move could be ignored when searching - have a one bit in the hash that contains rootDepth&1? And discard tt entries when it doesn't match (under conditions where GHI is more likely a problem)? I'm not sure whether there is a way to fit even one bit into the hash though. This has a problem too, when too little hash is overwritten, ie. when the search time is relativly short.
Another, possibly stupid idea. Don't store evals near zero in the tt.
Also, considering @Alayan-stk-2's "A difficulty is that GHI is very minor at fishtest conditions while the overhead is noticeable. In CCC conditions, GHI hurts more (though outright blunders that go into the PV are rare, the search tree has many), while the overhead would be less significant." would it be wise to include a solution even if it loses a few elo on fishtest?
Is there a list of positions with reproducible GHI issues? If we have a few instances it would be easier to evaluate possible solutions and would help with the decision whether to include it regardless of elo loss or not.
Sopel97
on 11 Dec 2019
@MichaelB7 Maybe you have time and want to give this a try?
This prevents the ttValue to be used once the 50-move counter reaches 90 and more, but still allows the ttMove to be retrieved. This seems to be less invasive than my 1st attempt.
joergoster
on 12 Dec 2019
Sure , when I get back home tomorrow night , I can test it.
MichaelB7
on 12 Dec 2019
@joergoster This looks very good.
First a snapshot that shows the game state ( 50 move rule count is being considered by the GUI - note the 91 reversible plies at the top)
https://www.dropbox.com/s/2543px6jd6nkdc7/Screen%20Shot%202019-12-13%20at%2010.00.50%20PM.png?dl=0
Kb4 is never considered:
dep score nodes time (not shown: tbhits knps seldep)
64 0.00 1.14G 0:21.93 Rb4 Rh1 Rb6 Kxh4 Rxg6
63 0.00 1.06G 0:20.47 Rb4 Rh1 Rf4 Kg3 Rf6 Kxh4 Rxg6
62 0.00 965.8M 0:18.82 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
61 0.00 870.9M 0:17.06 Rb4 Kg3 Ra4 Rh1 Ra6 Rxh4 Rxg6
60 0.00 801.0M 0:15.74 Rb4 Rh1 Rf4 Kg3 Rf6 Kxh4 Rxg6
59 0.00 720.2M 0:14.22 Rb4 Rc1+ Kd2 Rg1 Rc4 Rg4 Rc6 Kxh4 Rxg6
58 0.00 648.0M 0:12.84 Rb4 Kg3 Ra4 Rh1 Ra6 Rxh4 Rxg6
57 0.00 582.9M 0:11.62 Rb4 Rh1 Rb6 Kxh4 Rxg6
56 0.00 520.4M 0:10.46 Rb4 Rh1 Rb6 Kxh4 Rxg6
55 0.00 467.3M 0:09.45 Rb4 Rh1 Rf4 Kg3 Rf6 Rxh4 Rxg6
54 0.00 423.0M 0:08.60 Rb4 Rh1 Kd2 Kg3 Rb6 Rxh4 Rxg6
53 0.00 377.7M 0:07.73 Rb4 Rg1 Kd2 Rg4 Rb6 Rxh4 Rxg6
52 0.00 343.2M 0:07.05 Rb4 Rh1 Kd2 Kg3 Rb6 Rxh4 Rxg6
51 0.00 304.4M 0:06.30 Rb4 Rg1 Kd3 Rg4 Rb6 Kxh4 Rxg6
50 0.00 274.0M 0:05.68 Rb4 Rh1 Rb6 Kxh4 Rxg6
49 0.00 242.5M 0:05.07 Rb4 Rh1 Ra4 Kg3 Ra6 Rxh4 Rxg6
48 0.00 214.8M 0:04.52 Rb4 Rh1 Rd4 Kg3 Rd6 Rxh4 Rxg6
47 0.00 188.8M 0:04.01 Rb4 Rh1 Kd2 Kg3 Rb6 Rxh4 Rxg6
46 0.00 163.7M 0:03.50 Rb4 Rh1 Rb6 Kxh4 Rxg6
45 0.00 144.3M 0:03.11 Rb4 Rh1 Ra4 Kg3 Ra6 Rxh4 Rxg6
44 0.00 126.2M 0:02.74 Rb4 Rc1+ Kd2 Rh1 Rb6 Kxh4 Rxg6
43 0.00 107.4M 0:02.36 Rb4 Rh1 Re4 Kg3 Re6 Rxh4 Rxg6
42 0.00 90.7M 0:02.01 Rb4 Rh1 Rb6 Kxh4 Rxg6
41 0.00 77.6M 0:01.74 Rb4 Rh1 Rb6 Kxh4 Rxg6
40 0.00 66.6M 0:01.51 Rb4 Rh1 Kd2 Kg3 Rb6 Rxh4 Rxg6
39 0.00 55.7M 0:01.27 Rb4 Rh1 Kd2 Kg3 Rb6 Rxh4 Rxg6
38 0.00 47.2M 0:01.09 Rb4 Rg1 Rd4 Rg4 Rd6 Kxh4 Rxg6
37 0.00 37.6M 0:00.88 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
36 0.00 31.5M 0:00.74 Rb4 Rh1 Rf4 Kg3 Rf6 Rxh4 Rxg6
35 0.00 25.6M 0:00.62 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
34 0.00 21.6M 0:00.52 Rb4 Rh1 Rd4 Kg3 Rd6 Rxh4 Rxg6
33 0.00 17.6M 0:00.43 Rb4 Rh1 Kd2 Kg3 Rb6 Rxh4 Rxg6
32 0.00 14.3M 0:00.36 Rb4 Rh1 Kd2 Kg3 Rb6 Rxh4 Rxg6
31 0.00 11.8M 0:00.30 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
30 0.00 9.54M 0:00.24 Rb4 Rc1+ Kd2 Rg1 Rc4 Rg4 Rc6 Kxh4 Rxg6
29 0.00 7.59M 0:00.20 Rb4 Kg3 Ra4 Rh1 Ra6 Rxh4 Rxg6
28 0.00 5.97M 0:00.16 Rb4 Rh1 Ra4 Kg3 Ra6 Rxh4 Rxg6
27 0.00 4.76M 0:00.13 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
26 0.00 3.73M 0:00.10 Rb4 Rh1 Rb6 Kxh4 Rxg6
25 0.00 3.12M 0:00.09 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
24 0.00 2.41M 0:00.07 Rb4 Rh1 Rb6 Kxh4 Rxg6
23 0.00 1.96M 0:00.06 Rb4 Rh1 Rd4 Kg3 Rd6 Rxh4 Rxg6
22 0.00 1.59M 0:00.05 Rb4 Rh1 Rb6 Kxh4 Rxg6
21 0.00 1.29M 0:00.04 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
20 0.00 1.01M 0:00.03 Rb4 Rh1 Rd4 Kg3 Rd6 Rxh4 Rxg6
19 0.00 812336 0:00.02 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
18 0.00 657214 0:00.02 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
17 0.00 542838 0:00.02 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
16 0.00 446595 0:00.02 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
15 0.00 368025 0:00.01 Rb4 Rh1 Kd2 Kg3 Rb6 Rxh4 Rxg6
14 0.00 302290 0:00.01 Rb4 Rh1 Rb6 Kxh4 Rxg6
13 0.00 251983 0:00.01 Rb4 Rc1+ Kd2 Rh1 Rb6 Kxh4 Rxg6
12 0.00 185343 0:00.01 Rb4 Rg1 Rb6 Kxh4 Rxg6
11 0.00 131697 0:00.00 Rb4 Rg1 Kd3 Rg4 Rb6 Kxh4 Rxg6
10 0.00 98813 0:00.00 Rb4 Rc1+ Kd2 Rh1 Rb6 Kxh4 Rxg6
9 0.00 65453 0:00.00 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
8 0.00 46943 0:00.00 Rb4 Rh1 Kd3 Kg3 Rb6 Rxh4 Rxg6
7 0.00 28798 0:00.00 Rb4 Rc1+ Kd2 Rc8 Ke1 Re8+ Kf2 Rf8+ Ke1
6 0.00 17675 0:00.00 Rb4 Rc1+ Kd2 Rc8 Ke1 Rc1+
5 0.00 11717 0:00.00 Rb4 Rc1+ Kd2 Rc8 Rd4 Kg3
4 0.00 7006 0:00.00 Kc2 Rh1 Re6 Kxh4
3 0.00 4628 0:00.00 Kc2 Rh1 Re6
2 0.00 1136 0:00.00 Kb3 Ra1
1 0.00 233 0:00.00 Ra4
and when I move forward one more ( move 111. Kb4), in less than a 0.2 seconds, a tb mate is within site ( note this is a fully functional Stockfish equivalent with the one exception, when using 7 man EGTB , it is modified so that only one search is made with exactly 7 men EGTB, with more than 7 men, only 6 men EGTB are allowed to be searched since they are on SSD. This done to prevent nps dropping to a very slow level since most of my 7 men are on HDD. Also the score output is modified by a constant, just my own personal preference).
dep score nodes time (not shown: tbhits knps seldep)
59 +99.96 635.2M 0:19.19 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke4 Re1+ Kd3 Kb1 h4 Rd1+ Ke3 Rc1 h3 Rc3+ Kf4 Rxh3
58 +99.96 514.4M 0:15.60 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke4 Re1+ Kd3 Kb1 h4 Rd1+ Ke3 Rc1 h3 Rc3+
57 +99.96 267.4M 0:08.14 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke4 Re1+ Kd3 Kb1 h4 Rd1+ Ke4 Rd7 h3 Rh7 h2 Rh4+ Ke5 Rxh2
56 +99.96 240.1M 0:07.29 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke4 Re1+ Kd3 Kb1 h4 Rd1+ Ke4 Rd7 h3 Rh7 h2 Rh4+ Ke5 Rxh2
55 +99.96 236.8M 0:07.19 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke4 Re1+ Kd3 Kb1 h4 Rd1+ Ke4 Rd7 h3 Rh7 h2 Rh4+ Ke5 Rxh2
54 +99.96 208.6M 0:06.32 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke4 Re1+ Kd3 Kb1 h4 Rd1+ Ke4 Rd7 h3 Rh7 h2 Rh4+ Ke5 Rxh2
53 +99.96 149.7M 0:04.55 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke4 Re1+ Kd3 Kb1 h4 Rd1+ Ke4 Rd7 h3 Kc2
52 +99.96 119.0M 0:03.59 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke4 Re1+ Kd3 Kb1 h4 Rd1+
51 +99.96 64.2M 0:01.90 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Ra1 h2 Ra4+ Kh5 Ra1 h1=Q Rxh1+
50 +99.96 61.0M 0:01.80 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Ra1 h2 Ra4+ Kh5 Ra1 h1=Q Rxh1+
49 +99.96 58.6M 0:01.73 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Ra1 h2 Ra4+ Kh5 Ra1 h1=Q Rxh1+
48 +99.96 45.3M 0:01.32 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Ra1 h2 Ra4+ Kh5 Ra1 h1=Q Rxh1+
47 +99.96 42.7M 0:01.24 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Ra1 h2 Ra4+ Kh5 Ra1 h1=Q Rxh1+
46 +99.96 39.3M 0:01.14 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Ra1 h2 Ra4+ Kh5 Ra1 h1=Q Rxh1+
45 +99.96 37.9M 0:01.10 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Ra1 h2 Ra4+ Kh5 Ra1 h1=Q Rxh1+
44 +99.96 36.3M 0:01.05 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Ra1 h2 Ra4+ Kh5 Ra1 h1=Q Rxh1+
43 +99.96 31.6M 0:00.91 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Rh2 Rg2+ Rxg2+
42 +99.96 27.7M 0:00.79 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Rh2 Rg2+ Rxg2+
41 +99.96 26.6M 0:00.76 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kc2 g5 Rh1+ Kg4 Kb1 h4 Kb2 h3 Rh2 Rg2+ Rxg2+
40 +99.96 24.2M 0:00.69 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kb4 g5 Rh1+ Kg4 Kb5 h4 Kc6 h3 Rh2 Kh4 Kb6 g4 Kc7 Rf3 Kd6 g3 Rh1 g2 Ra1 h2 Ra4+ Kh5 Ra5+ Kg6 Rg5+ Kxg5 Ke5 g1=Q Ke6 h1=Q Ke5
39 +99.96 22.9M 0:00.65 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kb4 g5 Rh1+ Kg4 Kb5 h4 Kc6 h3 Rh2 Kh4 Kb6 g4 Kc7 Rf3 Kd6 g3 Ke6 gxh2 Ke7 h1=Q Kd6 Qd1+ Kc5 h2 Kc4
38 +99.96 22.1M 0:00.62 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Kb4 g5 Rh1+ Kg4 Kb5 h4 Kc6 h3 Rh2 Kh4 Kb6 g4 Kc7 Rf3 Kd6 g3 Ke6 gxh2 Ke7 h1=Q Ke6 Qe1+ Kd5 h2 Kc4 h1=Q Kd4
37 +99.96 21.2M 0:00.60 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Kc4 g5 Kxd3
36 +99.96 19.1M 0:00.53 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Kc4 g5 Kxd3
35 +99.96 16.9M 0:00.46 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Kc2 g3 Kxd3
34 +99.96 15.8M 0:00.42 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Kc2 g3 Kxd3
33 +99.96 15.0M 0:00.40 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Kc2 g3 Kxd3
32 +99.96 14.4M 0:00.38 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Kc2 g3 Kxd3
31 +99.96 13.9M 0:00.36 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Ka1 g3 Kb2 g2 Kc2 g1=R Kb2 Rg2+ Kc1 Rd1+ Kxd1
30 +99.96 13.4M 0:00.35 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Ka1 g3 Kb2 g2 Kc2 g1=R Kb2
29 +99.96 13.1M 0:00.34 Rc1 Rd4 Kg3 Re4 Rh1 Re6 Rxh4+ Kc3 Rg4 Re3+ Kh4 Re6 Rxg5 Re4+ Rg4 Kc2 Rxe4 Kd2 g5 Kc2 g4 Kd3 g3 Kc3 g2 Kc2
28 +99.96 12.1M 0:00.31 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Ka1 g3 Kb2 g2 Ka2
27 +99.96 11.7M 0:00.30 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Ka1 g3 Kb2 Kg4 Kb1 g2 Ka1 g1=Q+ Kb2
26 +99.96 11.3M 0:00.29 Rc1 Rd4 Kg3 Ka3 Rh1 Ka2 Rxh4 Rd6 Rg4 Rd3+ Kh4 Rd1 Rxg5 Kb3 Rg3+ Rd3 Rxd3+ Ka2 g5 Kb2 g4 Ka1 g3 Kb2 Kg4 Kb1 g2 Ka1 g1=Q+ Kb2
25 +99.96 11.1M 0:00.28 Rc1 Kb3 Kg3 Ra4 Rh1 Ra2 Kxh4 Rg2 Rh3+ Ka4 Rg3 Kb4 Rxg2 Kc4 Rxg5 Kb3 Rb5+ Kc3 g5 Kd4
24 +99.96 10.2M 0:00.25 Rc1 Kb5 Kg3 Kb4 Rh1 Kb5 Rxh4 Re5 Kf4 Rc5 Rg4 Kc6 Rxg5 Kb7 Rxc5 Kb6 Rc6+ Ka5 Ra6+ Kb5 h4 Kb4
23 +99.96 9.91M 0:00.24 Rc1 Kb5 Kg3 Kb6 Rh1 Ka7 Rxh4 Re1 Kf4 Rg1 Rg4 Kb8 Rxg1 Ka7 Rxg5 Kb6 Rg3 Kc5 g5 Kc6
22 +99.96 9.17M 0:00.22 Rc1 Kb5 Kg3 Ra4 Rh1 Kb6 Rxh4 Ra6 Kf4 Ra5 Rg4 Ra1 Kxg5 Rc1 h4 Rh1
21 +99.96 8.97M 0:00.21 Rc1 Kb5 Kg3 Ra4 Rh1 Kb6 Rxh4 Ra6 Kf4 Ra5 Rg4 Ra1 Kxg5 Rb1
20 +99.96 8.79M 0:00.21 Rc1 Kb5 Kg3 Ra4 Rh1 Kc5 Rxh4 Ra1 Kf4 Rg1 Rg4 Rg2 Rxg2 Kd5 Kxg5 Kd4
19 +99.96 8.53M 0:00.20 Rc1 Rd4 Kg3 Re4 Rh1 Kc5 Rxh4 Re1 Kf4 Rg1 Rg4 Rg2 Rxg2 Kd5 Kxg5 Ke6
18 +99.96 8.28M 0:00.19 Rc1 Kb3 Kg3 Rd4 Rh1 Re4 Rxh4 Re1 Kf4 Rg1 Rg4 Ra1 Rxg5 Ka4
17 +99.94 7.71M 0:00.18 Rc1 Kb3 Kg3 Ra4 Rh1 Kc2 Rxh4 Ra5 Rh2+ Kd3 Kh4 Ke3 Rg2 Kf4 Rg4+ Kf3 Rxg5 Rxg5
16 +7.89 5.56M 0:00.13 Rc1 Ka3 Kg3 Rd4 Rh1 Rc4 Rxh4 Rc5 Rf4 Rc2 h4 Ra2 h3 Rb2 h2 Rb3+ Kh4
15 +2.84 3.33M 0:00.08 Rc1 Kb5 Rh1 Kc5 Kg3 Kd6 Rxh4 Re5 Rd4+ Ke7 Kf4 Ke6 h4 Ra5 h3
14 +2.98 2.02M 0:00.05 Rc1 Kb5 Kg3 Ra4 Rh1 Kc6 Rxh4 Ra5 Rg4 Rb5 h4 Rc5 h3
13 +2.64 1.23M 0:00.03 Rc1 Kb5 Kg3 Rb4 Rh1 Kc6 Rxh4 Rb5 Rg4 Kd7 h4 Ke6 h3
12 +1.55 674442 0:00.02 Rc1 Kb5 Kg3 Ra4 Rh1 Rb4 Rxh4 Rb1 Rg4
11 +1.24 434503 0:00.01 Rc1 Kb5 Kg3 Rd4 Rh1 Re4 Rxh4 Re6 Rf4
10 +0.99 280945 0:00.01 Rc1 Kb3 Kg3 Kb2 Rh1 Rb4 Rxh4 Rb3+ Kf4
9 +0.88 177649 0:00.01 Rc1 Rd4 Kg3 Kb3 Rh1 Rd6 Rxh4
8 +0.72 98447 0:00.00 Rc1 Kb5 Kg3 Rb4 Rh1 Kc6 Rxh4 Rb3+ Kg4
7 +0.60 47270 0:00.00 Rc1 Kb5 Kg3 Rd4 Rh1
6 +0.33 33135 0:00.00 Rc1 Rf4 Kg3
5 0.00 12162 0:00.00 Rh1 Re6 Kxh4 Rxg6
4 0.00 7928 0:00.00 Rh1 Re6 Kxh4 Rxg6
3 0.00 2964 0:00.00 Rh1 Re6 Kxh4
2 +0.17 951 0:00.00 Rh1 Kc4 Kg3
1 0.00 108 0:00.00 Rh1
To recap the methodology, load the game from the pgn file, fast forward to move 50. At move 50, move the game forward , allowing each position to be analyze for 6 seconds, then move forward one ply, wait 6 seconds , move forward again. wait 6 seconds - up to move 111. The point of the waiting is to allow the hash to get filled as if SF was playing a game. The hash was set at 16 GB, xBoard's maximum hash setting.
As follow up on the nps, they gets back to normal when there are 8 men left, good to know. So searching with 8 men is ok on HDD.
MichaelB7
on 14 Dec 2019
Here I test it with true stockfish with your modifications, using the methodology above, after a while , the nps picks up as the probes must be cached, SF sees a tb mate in 0.01 second
dep score nodes time (not shown: tbhits knps seldep)
52 +149.04 446.7M 0:15.03 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Rg2 h3 Rg3+ Ke4 Rxg4+
51 +149.04 26.9M 0:00.92 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Kd2 Kd4 h2 Rd3+ Ke2 Re3+ Kf2 Re4 g2 Rf4+ Ke1 Kc3 g1=Q Re4+ Qe3+ Rxe3+
50 +149.04 24.6M 0:00.84 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Kd2 Kd4 h2 Rd3+ Ke2 Re3+ Kf2 Re4 g2 Rf4+ Ke1 Kc3 g1=Q Rf3 Qe3+ Rxe3+
49 +149.04 22.4M 0:00.77 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Kd2 Kd4 h2 Rd3+ Ke2 Re3+ Kf2 Re4 g2 Rf4+ Ke1 Kc3 g1=Q Ra4 Qc5+ Kd3 Qc2+ Kxc2
48 +149.04 21.2M 0:00.73 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Kd2 Kd4 h2 Rd3+ Ke2 Re3+ Kf2 Re4 g2 Rf4+
47 +149.04 12.5M 0:00.42 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Kd2 Rc4 h2 Kd6 g2 Rd4+ Kc3 Re4 g1=Q Kd5
46 +149.04 10.4M 0:00.35 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Kd2 Rc4 h2 Kd6 g2 Ke7 g1=R Rd4+
45 +149.04 9.52M 0:00.32 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Kd2 Kb4 g2 Rc5 g1=Q Rd5+ Qd4+ Rxd4+
44 +149.04 8.03M 0:00.27 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Ke4 Rc2 g2 Re2+ Kf3 Re6 g1=Q+ Kd6 h2 Rf6+
43 +149.04 7.36M 0:00.24 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Ke4 Rc2 g2 Re2+ Kf3 Re6 g1=Q+ Kd6 h2 Rf6+
42 +149.04 5.75M 0:00.18 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Kd5 h2 Re4+ Kf3 Rd4 g3 Rd3+ Kg4 Rd4+ Kh5 Ke4 g2 Rd5+ Kh4 Kf4 g1=Q Rd6
41 +149.04 5.27M 0:00.16 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Rc4 g3 Rc3+ Kf4 Kb4 g2 Rc4+
40 +149.04 4.90M 0:00.15 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kd2 Rd4+ Ke3 Kd5 h2 Re4+ Kf3 Rd4 g3 Rd3+
39 +149.04 4.67M 0:00.14 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke1 Rd4 h3 Re4+ Kf2 Rf4+ Kg2 Kd5 h2 Kc5 g3 Kb6 Rc1 Ra4 h1=Q Ka5
38 +149.04 3.44M 0:00.10 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Ke6 g1=Q Kd7
37 +149.04 3.14M 0:00.09 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Ke8 g1=Q Kd8
36 +149.04 2.92M 0:00.08 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7 g1=Q+ Kf8
35 +149.04 2.68M 0:00.07 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7 g1=Q+ Kh7
34 +149.04 2.48M 0:00.07 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7 g1=Q+ Kh7
33 +149.04 2.29M 0:00.06 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7 g1=Q+ Kh7
32 +149.04 2.13M 0:00.06 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7 g1=Q+ Kh6
31 +149.04 1.97M 0:00.05 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7 g1=Q+ Kh6
30 +149.04 1.82M 0:00.05 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7 g1=Q+ Kh6
29 +149.04 1.69M 0:00.04 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7 g1=R+ Kh6
28 +149.04 1.54M 0:00.04 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7
27 +149.04 1.43M 0:00.04 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7 g2 Kg7
26 +149.04 1.31M 0:00.03 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6 g3 Kf7
25 +149.04 1.21M 0:00.03 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6
24 +149.04 1.12M 0:00.03 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6
23 +149.04 1.03M 0:00.03 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5 h2 Ke6
22 +149.04 954458 0:00.02 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5
21 +149.04 867239 0:00.02 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+ Kf2 Rd2+ Ke3 Kc6 Kxd2 Kd5 h3 Ke5
20 +149.04 791020 0:00.02 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+
19 +149.04 727272 0:00.02 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+
18 +149.04 663207 0:00.02 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+
17 +149.04 606622 0:00.02 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+
16 +149.04 551785 0:00.02 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+ Kg3 Kc5 g4 Rd6 h4 Rd3+
15 +149.04 498535 0:00.01 Rc1 Kb5 Kg3 Ra4 Rh1 Rf4 Kxf4 Kc4 Rxh4 Kc5 Kxg5 Kb5 Rg4 Kc6 Kh4 Kd7
14 +149.04 448172 0:00.01 Rc1 Kb5 Kg3 Ra4 Rh1 Rf4 Kxf4 Kc4 Rxh4 Kc5 Kxg5 Kd5
13 +149.04 407663 0:00.01 Rc1 Kb5 Kg3 Ra4 Rh1 Rf4 Kxf4 Kc4 Rxh4 Kc5 Kxg5 Kb6
12 +149.04 372233 0:00.01 Rc1 Rf4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4 Re6 g5 Rf6+
11 +149.04 334460 0:00.01 Rc1 Kb5 Kg3 Ra4 Rh1 Rf4 Kxf4 Kc4 Rxh4 Kc5 Kxg5 Kc6
10 +149.04 301788 0:00.01 Rc1 Kb5 Kg3 Ra4 Rh1 Rf4 Kxf4 Kc4 Rxh4 Kc5
9 +3.27 118895 0:00.00 Rc1 Rd4 Kg3 Re4 Rh1 Re3+ Kxh4 Re4+ Kxg5 Re5+ Kf4
8 +0.85 76768 0:00.00 Rc1 Ka4 Rh1 Rc4 Kg3
7 +0.36 52865 0:00.00 Rc1 Kb5 Kg3 Kb6 Rh1
6 +0.50 16326 0:00.00 Rc1 Kb5 Kg3 Kb6 Rc2
5 +0.55 10187 0:00.00 Rc1 Kb3 Kg3 Kb4 Rb1+ Kc3
4 +0.53 4711 0:00.00 Rc1 Kb5 Kg3
3 +0.47 2532 0:00.00 Rc1 Kb5 Kg3
2 +1.09 835 0:00.00 Rh1 Rf4
1 +0.23 109 0:00.00 Rc1
I retested stockfish without @joergoster patch after 111.Kb4 above and here are the results:
It took a depth of 231 ( 3.96 seconds ) to Kb4 is no longer a draw , which is more than the time used by SF in the game. It is repeatable without the patch.
dep score nodes time (not shown: tbhits knps seldep)
231 +147.92! 160.5M 0:04.28 Rc1!
231 +30.52! 160.1M 0:04.27 Rc1!
231 +24.08! 159.7M 0:04.26 Rc1!
231 +18.95! 159.5M 0:04.25 Rc1!
231 +14.86! 159.3M 0:04.24 Rc1!
231 +11.61! 159.1M 0:04.23 Rc1!
231 +9.02! 158.9M 0:04.23 Rc1!
231 +6.97! 158.7M 0:04.22 Rc1!
231 +5.34! 158.2M 0:04.20 Rc1!
231 +4.06! 157.3M 0:04.18 Rc1!
231 +3.05! 156.7M 0:04.16 Rc1!
231 +2.26! 155.9M 0:04.14 Rc1!
231 +1.65! 155.4M 0:04.12 Rc1!
231 +1.17! 154.5M 0:04.10 Rc1!
231 +0.81! 154.1M 0:04.09 Rc1!
231 +0.54! 153.8M 0:04.08 Rc1!
231 +0.34! 152.2M 0:04.03 Rc1!
231 +0.19! 151.1M 0:04.00 Rc1!
231 +0.09! 150.0M 0:03.96 Rc1!
230 0.00 142.4M 0:03.75 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
229 0.00 138.0M 0:03.64 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
228 0.00 135.3M 0:03.57 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
227 0.00 132.9M 0:03.50 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
226 0.00 130.8M 0:03.45 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
225 0.00 128.8M 0:03.39 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
224 0.00 126.7M 0:03.34 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
223 0.00 124.5M 0:03.28 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
222 0.00 122.5M 0:03.23 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
221 0.00 120.7M 0:03.18 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
220 0.00 116.2M 0:03.07 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
219 0.00 114.1M 0:03.02 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
218 0.00 110.0M 0:02.92 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
217 0.00 105.8M 0:02.80 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
216 0.00 103.7M 0:02.74 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
215 0.00 101.8M 0:02.69 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
214 0.00 99.9M 0:02.64 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
213 0.00 98.2M 0:02.59 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
212 0.00 96.5M 0:02.54 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
211 0.00 94.5M 0:02.49 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
210 0.00 92.3M 0:02.43 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
209 0.00 89.9M 0:02.37 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
208 0.00 88.1M 0:02.33 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
207 0.00 86.2M 0:02.28 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
206 0.00 84.4M 0:02.23 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
205 0.00 82.5M 0:02.18 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
204 0.00 80.4M 0:02.13 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
203 0.00 78.5M 0:02.08 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
202 0.00 76.7M 0:02.03 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
201 0.00 74.6M 0:01.97 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
200 0.00 72.6M 0:01.92 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
199 0.00 71.0M 0:01.87 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
198 0.00 68.8M 0:01.81 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
197 0.00 66.9M 0:01.76 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
196 0.00 65.1M 0:01.71 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
195 0.00 63.8M 0:01.68 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
194 0.00 62.4M 0:01.64 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
193 0.00 61.0M 0:01.60 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
192 0.00 59.5M 0:01.57 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
191 0.00 57.7M 0:01.52 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
190 0.00 56.1M 0:01.48 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
189 0.00 54.6M 0:01.44 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
188 0.00 53.2M 0:01.40 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
187 0.00 50.7M 0:01.34 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
186 0.00 49.0M 0:01.29 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
185 0.00 47.0M 0:01.24 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
184 0.00 44.9M 0:01.19 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
183 0.00 43.1M 0:01.14 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
182 0.00 41.7M 0:01.10 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
181 0.00 40.0M 0:01.06 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
180 0.00 38.3M 0:01.01 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
179 0.00 36.6M 0:00.97 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
178 0.00 35.0M 0:00.93 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
177 0.00 33.4M 0:00.89 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
176 0.00 32.1M 0:00.86 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
175 0.00 30.9M 0:00.82 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
174 0.00 29.5M 0:00.79 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
173 0.00 28.1M 0:00.75 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
172 0.00 26.8M 0:00.72 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
171 0.00 25.5M 0:00.69 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
170 0.00 24.0M 0:00.65 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
169 0.00 22.8M 0:00.62 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
168 0.00 21.7M 0:00.58 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
167 0.00 20.7M 0:00.55 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
166 0.00 19.8M 0:00.53 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
165 0.00 19.0M 0:00.50 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
164 0.00 17.9M 0:00.47 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
163 0.00 17.2M 0:00.45 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
162 0.00 16.4M 0:00.43 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
161 0.00 15.4M 0:00.40 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
160 0.00 14.7M 0:00.38 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
159 0.00 13.9M 0:00.36 Rc1 Kb3 Rg1 Kc3 Rg4 Re6 Kxh4
158 0.00 13.4M 0:00.34 Kg3 Kc3
157 0.00 13.0M 0:00.33 Kg3
156 0.00 12.3M 0:00.32 Kg3 Kc3
155 0.00 12.2M 0:00.32 Kg3 Re6 Kxh4
154 0.00 12.2M 0:00.31 Kg3
153 0.00 12.1M 0:00.31 Kg3
152 0.00 11.9M 0:00.31 Kg3 Re6 Kxh4
151 0.00 11.9M 0:00.30 Kg3
150 0.00 11.5M 0:00.30 Kg3 Re6 Kxh4
149 0.00 11.2M 0:00.29 Kg3 Re6 Kxh4
148 0.00 11.1M 0:00.28 Kg3
147 0.00 10.8M 0:00.28 Kg3 Kc3
146 0.00 10.5M 0:00.27 Kg3 Re6 Kxh4
145 0.00 10.4M 0:00.27 Kg3 Re6 Kxh4
144 0.00 10.4M 0:00.27 Kg3 Re6 Kxh4
143 0.00 10.2M 0:00.26 Kg3 Kc3
142 0.00 10.1M 0:00.26 Kg3
141 0.00 9.85M 0:00.26 Kg3
140 0.00 9.54M 0:00.25 Kg3
139 0.00 9.48M 0:00.25 Kg3 Kc3
138 0.00 9.41M 0:00.24 Kg3 Re6 Kxh4
137 0.00 9.14M 0:00.24 Kg3 Kc3
136 0.00 9.08M 0:00.24 Kg3 Kc3
135 0.00 9.01M 0:00.23 Kg3
134 0.00 8.94M 0:00.23 Kg3
133 0.00 8.87M 0:00.23 Kg3 Re6 Kxh4
132 0.00 8.36M 0:00.22 Kg3
131 0.00 7.85M 0:00.20 Kg3
130 0.00 7.56M 0:00.20 Kg3
129 0.00 7.38M 0:00.19 Kg3
128 0.00 7.22M 0:00.19 Kg3
127 0.00 7.16M 0:00.19 Kg3
126 0.00 7.10M 0:00.19 Kg3 Kc3
125 0.00 6.92M 0:00.18 Kg3 Re6 Kxh4
124 0.00 6.79M 0:00.18 Kg3 Re6 Kxh4
123 0.00 6.53M 0:00.17 Kg3 Re6 Kxh4
122 0.00 6.38M 0:00.17 Kg3 Re6 Kxh4
121 0.00 6.26M 0:00.16 Kg3 Re6 Kxh4
120 0.00 6.20M 0:00.16 Kg3
119 0.00 6.15M 0:00.16 Kg3
118 0.00 5.98M 0:00.16 Kg3
117 0.00 5.91M 0:00.16 Kg3
116 0.00 5.85M 0:00.15 Kg3
115 0.00 5.76M 0:00.15 Kg3 Re6 Kxh4
114 0.00 5.57M 0:00.15 Kg3 Kc3
113 0.00 5.40M 0:00.14 Kg3 Re6 Kxh4
112 0.00 5.38M 0:00.14 Kg3 Re6 Kxh4
111 0.00 5.19M 0:00.14 Kg3
110 0.00 4.88M 0:00.13 Kg3 Re6 Kxh4
109 0.00 4.75M 0:00.13 Kg3
108 0.00 4.57M 0:00.12 Kg3 Re6 Kxh4
107 0.00 4.55M 0:00.12 Kg3
106 0.00 4.50M 0:00.12 Kg3
105 0.00 4.44M 0:00.12 Kg3 Kc3
104 0.00 4.38M 0:00.12 Kg3 Kc3
103 0.00 4.32M 0:00.12 Kg3 Re6 Kxh4
102 0.00 4.19M 0:00.11 Kg3 Kc3
101 0.00 4.12M 0:00.11 Kg3 Re6 Kxh4
100 0.00 4.01M 0:00.11 Kg3
99 0.00 3.88M 0:00.11 Kg3 Re6 Kxh4
98 0.00 3.68M 0:00.10 Kg3
97 0.00 3.58M 0:00.10 Kg3
96 0.00 3.49M 0:00.10 Kg3 Re6 Kxh4
95 0.00 3.48M 0:00.10 Kg3 Kc3
94 0.00 3.42M 0:00.09 Kg3 Re6 Kxh4
93 0.00 3.41M 0:00.09 Kg3 Re6 Kxh4
92 0.00 3.21M 0:00.09 Kg3
91 0.00 2.97M 0:00.08 Kg3 Re6 Kxh4
90 0.00 2.89M 0:00.08 Kg3
89 0.00 2.84M 0:00.08 Kg3
88 0.00 2.74M 0:00.08 Kg3
87 0.00 2.61M 0:00.07 Kg3 Re6 Kxh4
86 0.00 2.59M 0:00.07 Kg3
85 0.00 2.56M 0:00.07 Kg3
84 0.00 2.46M 0:00.07 Kg3
83 0.00 2.42M 0:00.07 Kg3 Re6 Kxh4
82 0.00 2.34M 0:00.07 Kg3 Kc3
81 0.00 2.29M 0:00.07 Kg3 Kc3
80 0.00 2.24M 0:00.06 Kg3 Re6 Kxh4
79 0.00 2.16M 0:00.06 Kg3 Kc3
78 0.00 2.12M 0:00.06 Kg3
77 0.00 2.08M 0:00.06 Kg3
76 0.00 2.05M 0:00.06 Kg3 Kc3
75 0.00 1.95M 0:00.06 Kg3
74 0.00 1.91M 0:00.06 Kg3 Re6 Kxh4
73 0.00 1.90M 0:00.06 Kg3
72 0.00 1.88M 0:00.06 Kg3
71 0.00 1.85M 0:00.05 Kg3
70 0.00 1.82M 0:00.05 Kg3
69 0.00 1.79M 0:00.05 Kg3
68 0.00 1.75M 0:00.05 Kg3 Re6 Kxh4
67 0.00 1.66M 0:00.05 Kg3 Kc3
66 0.00 1.64M 0:00.05 Kg3
65 0.00 1.62M 0:00.05 Kg3 Kc3
64 0.00 1.59M 0:00.05 Kg3
63 0.00 1.53M 0:00.05 Kg3
62 0.00 1.49M 0:00.04 Kg3 Re6 Kxh4
61 0.00 1.41M 0:00.04 Kg3 Kc3
60 0.00 1.30M 0:00.04 Kg3
59 0.00 1.28M 0:00.04 Kg3 Re6 Kxh4
58 0.00 1.26M 0:00.04 Kg3 Re6 Kxh4
57 0.00 1.20M 0:00.04 Kg3 Kc3
56 0.00 1.04M 0:00.03 Kg3 Re6 Kxh4
55 0.00 1.04M 0:00.03 Kg3 Re6 Kxh4
54 0.00 1.03M 0:00.03 Kg3 Kc3
53 0.00 716178 0:00.02 Kg3
52 0.00 692109 0:00.02 Kg3 Re6 Kxh4
51 0.00 660922 0:00.02 Kg3 Kc3
50 0.00 614183 0:00.02 Kg3 Re6 Kxh4
49 0.00 568973 0:00.02 Kg3
48 0.00 536720 0:00.02 Kg3 Re6 Kxh4
47 0.00 527465 0:00.02 Kg3 Re6 Kxh4
46 0.00 518910 0:00.02 Kg3 Re6 Kxh4
45 0.00 510724 0:00.02 Kg3
44 0.00 496707 0:00.02 Kg3 Re6 Kxh4
43 0.00 458538 0:00.02 Kg3
42 0.00 434767 0:00.02 Kg3 Re6 Kxh4
41 0.00 429907 0:00.02 Kg3 Re6 Kxh4
40 0.00 389963 0:00.01 Kg3 Kc3
39 0.00 372832 0:00.01 Kg3 Re6 Kxh4
38 0.00 367196 0:00.01 Kg3 Re6 Kxh4
37 0.00 361962 0:00.01 Kg3
36 0.00 355023 0:00.01 Kg3 Re6 Kxh4
35 0.00 347288 0:00.01 Kg3
34 0.00 340632 0:00.01 Kg3 Re6 Kxh4
33 0.00 335784 0:00.01 Kg3 Kc3
32 0.00 308037 0:00.01 Kg3
31 0.00 294108 0:00.01 Kg3
30 0.00 285873 0:00.01 Kg3 Kc3
29 0.00 139952 0:00.01 Kg3
28 0.00 130682 0:00.01 Kg3
27 0.00 123842 0:00.01 Kg3
26 0.00 118969 0:00.01 Kg3
25 0.00 116133 0:00.01 Kg3 Re6 Kxh4
24 0.00 110570 0:00.01 Kg3 Re6 Kxh4
23 0.00 105728 0:00.00 Kg3 Re6 Kxh4
22 0.00 92332 0:00.00 Kg3
21 0.00 85533 0:00.00 Kg3
20 0.00 81879 0:00.00 Kg3 Re6 Kxh4
19 0.00 78230 0:00.00 Kg3 Kc3
18 0.00 72557 0:00.00 Kg3
17 0.00 69313 0:00.00 Kg3
16 0.00 66528 0:00.00 Kg3 Re6 Kxh4
15 0.00 61226 0:00.00 Kg3
14 0.00 57514 0:00.00 Kg3
13 0.00 52523 0:00.00 Kg3
12 0.00 50978 0:00.00 Kg3
11 0.00 49485 0:00.00 Kg3
10 0.00 47916 0:00.00 Kg3
9 0.00 46323 0:00.00 Kg3 Re6 Kxh4
8 0.00 44182 0:00.00 Kg3
7 0.00 41746 0:00.00 Kg3
6 0.00 40255 0:00.00 Kg3
5 0.00 38880 0:00.00 Kg3
4 0.00 37735 0:00.00 Kg3 Kc3
3 0.00 36261 0:00.00 Kg3
2 0.00 34068 0:00.00 Rc1 Kb3
1 +0.57 119 0:00.00 Rc1
@MichaelB7 Thank you very much for testing! This looks really good.
I will open a PR later.
joergoster
on 14 Dec 2019
Just wondering if we can create a list of positions where this happened, following seem related:
https://github.com/official-stockfish/Stockfish/issues/2138
https://github.com/official-stockfish/Stockfish/issues/501
none seem to be easy to reproduce (as in reproducible with a defined sequence of UCI commands)
vondele
on 15 Dec 2019
It happened again in CCC finals game 119
adentong
on 22 Dec 2019
PGN of game 119:
[Event "CCC11 Finals! (30|5)"]
[Site "?"]
[Date "2019.12.22"]
[Round "1"]
[White "Stockfish"]
[Black "Lc0"]
[Result "1/2-1/2"]
[ECO "B10"]
[GameDuration "01:16:43"]
[GameEndTime "2019-12-22T01:19:14.851 PST"]
[GameStartTime "2019-12-22T00:02:31.795 PST"]
[Opening "Caro-Kann defense"]
[PlyCount "205"]
[TimeControl "1800+5"]
1. e4 c6 2. Nf3 d5 3. d4 g6 4. e5 Nh6 5. Bf4 Nf5 6. c3 Qb6 7. Qc1 h5 8. Nbd2 Bh6
9. Bxh6 Nxh6 10. Be2 Nd7 11. O-O Nf8 12. b4 a5 13. bxa5 Qxa5 14. a4 Qd8 15. c4
Ne6 16. Nb3 Kf8 17. cxd5 Qxd5 18. Bc4 Qd8 19. a5 Kg7 20. Rd1 Nc7 21. Nc5 Nd5 22.
Bxd5 cxd5 23. Qc3 b6 24. axb6 Rxa1 25. Qxa1 Qxb6 26. Rb1 Qc7 27. h3 Nf5 28. Ng5
h4 29. Rb4 Rf8 30. Nf3 Rh8 31. Kh2 Rg8 32. Ng5 Rh8 33. Rb1 Rg8 34. g4 hxg3+ 35.
fxg3 Rf8 36. g4 Ne3 37. Qa2 Nc4 38. Rb7 Qa5 39. Qxa5 Nxa5 40. Rxe7 Nc6 41. Rc7
Nxd4 42. e6 Nb5 43. Re7 Nd6 44. h4 d4 45. Ra7 Nb5 46. Ra5 Nc7 47. exf7 Bxg4 48.
Ra7 Nb5 49. Nce6+ Bxe6 50. Nxe6+ Kf6 51. Ra6 Rxf7 52. Nxd4+ Kg7 53. Nxb5 Kh6 54.
Kg3 Rf1 55. Nd4 Rg1+ 56. Kf4 Kh5 57. Nf3 Rg4+ 58. Ke3 Rb4 59. Kf2 Rg4 60. Rb6
Ra4 61. Rd6 Rg4 62. Rb6 Ra4 63. Rb5+ Kh6 64. Ke2 Ra2+ 65. Kd3 Ra3+ 66. Ke2 Ra2+
67. Kd3 Ra3+ 68. Ke4 Ra4+ 69. Nd4 Rc4 70. Ra5 Rb4 71. Rc5 Ra4 72. Ke3 Rb4 73.
Ra5 Rc4 74. Rd5 Ra4 75. Rb5 Ra1 76. Rd5 Ra4 77. Re5 Ra3+ 78. Kf4 Ra4 79. Ke3
Ra3+ 80. Ke2 Ra4 81. Rd5 Ra1 82. Kd2 Ra2+ 83. Ke1 Rh2 84. Nf3 Rg2 85. Kf1 Ra2
86. Kg1 Rb2 87. Kf1 Ra2 88. Re5 Rb2 89. Rc5 Ra2 90. Rd5 Rb2 91. Nd2 Rb4 92. Nf3
Rb2 93. Rg5 Rb4 94. Rc5 Ra4 95. Re5 Ra2 96. Rb5 Ra4 97. Kf2 Ra2+ 98. Kg3 Ra3 99.
Rb7 Rc3 100. Kg4 Rc4+ 101. Kg3 Rc3 102. Kg4 Rc4+ 103. Kg3 1/2-1/2
Just a personal note: whatever the reason for blundering a win or a draw is, I don't like incomplete TB sets are being used. Period.
joergoster
on 22 Dec 2019
The engine should not blunder because of partial 7-men.
Alayan-stk-2
on 22 Dec 2019
@Alayan-stk-2 That's right.
But it will eventually give the impression of a bug to some if one endgame is being played perfectly, while another one looks like blundered away.
joergoster
on 26 Dec 2019
Maybe the following is interesting and might help discover "the problem".
Tested with default one core and hash 2048
White to move and win
7k/Q7/n1P4q/3bpp2/3n4/2p5/1pP3RP/7K w - - 11 1
White will keep checking with his Queen, take all black's pieces and pawns in the process.
Latest SF (20200108) see all this, and very quickly output +20 scores.
However, if you clearhash, and set the board after the Qa7-a8+ move
Q6k/8/n1P4q/3bpp2/3n4/2p5/1pP3RP/7K b - - 12 1
SF stays at 0.00, and start to increase scores only at depth 55/67+
Rocky640
on 9 Jan 2020
Closing this issue, since bae019b53e5c2bfcf0d69b4ebfc52b4f4de762eb has been merged. Thanks!
snicolet
on 9 Jan 2020
@Rocky640
Ooops, sorry, I had not seen that your last comment was just a few minutes old when I closed the issue! :-(
snicolet
on 9 Jan 2020
it is not related to the 50 move rule, more probably to 3-fold but quite odd anyway !
Rocky640
on 9 Jan 2020
with the #2453 merged, the effect is indeed similar
Tested with default one core and hash 2048
7k/Q7/n1P4q/3bpp2/3n4/2p5/1pP3RP/7K w - - 11 1
After clearhash, and set the board after the Qa7-a8+ move
Q6k/8/n1P4q/3bpp2/3n4/2p5/1pP3RP/7K b - - 12 1
BUT it seem also that larger hash values produce another strange behaviour.
Tested with default one core and hash 4096
7k/Q7/n1P4q/3bpp2/3n4/2p5/1pP3RP/7K w - - 11 1
After clearhash, and set the board after the Qa7-a8+ move
Q6k/8/n1P4q/3bpp2/3n4/2p5/1pP3RP/7K b - - 12 1
The mate core is lost for depth 54-101- !!!!
Rocky640
on 9 Jan 2020
Related issues
nguyenpham
·
4Comments
anonymous7002
·
3Comments
Technologov
·
3Comments
Silver-Fang
·
7Comments
niklasf
·
5Comments