New engine releases 2020

[Kieren Pearson]

That’s a really good point, I’ll take down the release for now and try to compile a static exe in the meantime

[Kieren Pearson]

Halogen 6 released

https://github.com/KierenP/Halogen/releases/tag/v6

In single threaded, probably about +120 ELO (2463 CCRL Blitz)
In multi threaded, probably about +180 ELO (2587 CCRL Blitz)

Both popcnt and non-popcnt builts supplied. Can anyone who uses non-popcnt builds verify it works on their machine? I have no way of making sure

Thanks for the new release. I tested the popcount compile and noticed it is not a static one.
Thus people would need to dig out the missing and fitting libraries.
Better create a static one for those, who cant compile and still use old hardware.

This had now been fixed. I actually exposed a gcc bug with gcc version 8.1.0 but found this workaround.

https://stackoverflow.com/questions/100 ... lto-option

Please let me know if there are any issues.

[Guenther]

Thanks for the new release. I tested the popcount compile and noticed it is not a static one.
Thus people would need to dig out the missing and fitting libraries.
Better create a static one for those, who cant compile and still use old hardware.

This had now been fixed. I actually exposed a gcc bug with gcc version 8.1.0 but found this workaround.

https://stackoverflow.com/questions/100 ... lto-option

Please let me know if there are any issues.

Running inside WB with 512MB hash (my compilation from 2020-04):

Code: Select all

dep	score	nodes	time	(not shown:  tbhits	knps	seldep)
 17	+0.23 	47.0M  	1:11.39	g1f3 g8f6 e2e3 d7d5 d2d4 e7e6 f1d3 f8d6 e1g1 e8g8 c2c4 d5c4 d3c4 b8c6 b1c3 c8d7 c1d2 string thread 3
 16	+0.22 	22.7M  	0:34.84	g1f3 g8f6 e2e3 d7d5 d2d4 e7e6 f1d3 f8d6 e1g1 e8g8 c2c4 c7c5 c4d5 e6d5 b1c3 b8c6 string thread 3
 15	+0.23 	14.1M  	0:21.91	g1f3 g8f6 e2e3 d7d5 d2d4 e7e6 f1d3 f8d6 e1g1 b8c6 c2c4 d5c4 d3c4 e8g8 b1c3 string thread 3
 14	+0.18 	8.58M  	0:13.52	g1f3 g8f6 e2e3 d7d5 b1c3 e7e6 d2d4 b8c6 f1d3 f8d6 e1g1 e8g8 c1d2 c8d7 string thread 3
 13	+0.14 	5.48M  	0:08.85	g1f3 g8f6 e2e3 d7d5 f1e2 e7e6 c2c4 f8e7 c4d5 e6d5 e1g1 e8g8 d2d4 string thread 3
 12	+0.18 	3.47M  	0:05.63	g1f3 g8f6 e2e3 d7d5 b1c3 e7e6 d2d4 f8d6 f1d3 e8g8 e1g1 b8c6 string thread 3
 11	+0.29 	2.54M  	0:04.14	g1f3 g8f6 e2e3 d7d5 b1c3 string thread 3
 10	+0.18 	779477	0:01.36	e2e4 b8c6 b1c3 g8f6 g1f3 e7e5 f1d3 f8d6 e1g1 e8g8 string thread 3
  9	+0.31 	227495	0:00.43	g1f3 g8f6 e2e3 e7e6 b1c3 f8d6 f1d3 e8g8 e1g1 string thread 3
  8	+0.31 	93606  	0:00.18	g1f3 g8f6 e2e3 b8c6 f1d3 d7d5 e1g1 e7e5 string thread 3
  7	+0.38 	42338  	0:00.09	g1f3 g8f6 e2e3 b8c6 f1d3 c6b4 d3h7 string thread 3
  6	+0.18 	15638  	0:00.04	g1f3 g8f6 b1c3 b8c6 d2d4 d7d5 string thread 3
  5	+0.25 	5148    	0:00.02	g1f3 g8f6 b1c3 b8c6 d2d4 string thread 3
  4	+0.18 	1550    	0:00.01	g1f3 g8f6 b1c3 b8c6 string thread 3
  3	+0.33 	728      	0:00.01	g1f3 g8f6 b1c3 string thread 3
  2	+0.18 	116      	0:00.01	g1f3 g8f6 string thread 3
  1	+0.37 	21        	0:00.00	g1f3 string thread 3

From cmd with 512MB hash after I figured out it does not understand go time x and go depth x but go infinite ;-)

Code: Select all

Halogen 6 x64
uci
id name Halogen 6
id author Kieren Pearson
option name Clear Hash type button
option name Hash type spin default 2 min 2 max 8192
option name Threads type spin default 1 min 1 max 8
uciok
ucinewgame
isready
readyok
position startpos
setoption name Hash value 512
go infinite
info depth 1 seldepth 1 score cp 37 time 0 nodes 21 nps 21000 hashfull 0 hashHitRate 0 pawnHitRate 346 pv g1f3 string thread 3
info depth 2 seldepth 2 score cp 18 time 10 nodes 116 nps 11000 hashfull 0 hashHitRate 0 pawnHitRate 468 pv g1f3 g8f6 string thread 3
info depth 3 seldepth 3 score cp 33 time 20 nodes 728 nps 36000 hashfull 0 hashHitRate 0 pawnHitRate 600 pv g1f3 g8f6 b1c3 string thread 3
info depth 4 seldepth 4 score cp 18 time 30 nodes 1550 nps 51000 hashfull 0 hashHitRate 21 pawnHitRate 662 pv g1f3 g8f6 b1c3 b8c6 string thread 3
info depth 5 seldepth 5 score cp 25 time 50 nodes 5148 nps 102000 hashfull 0 hashHitRate 17 pawnHitRate 793 pv g1f3 g8f6 b1c3 b8c6 d2d4 string thread 3
info depth 6 seldepth 6 score cp 18 time 80 nodes 15638 nps 195000 hashfull 0 hashHitRate 22 pawnHitRate 812 pv g1f3 g8f6 b1c3 b8c6 d2d4 d7d5 string thread 3
info depth 7 seldepth 7 score cp 38 time 150 nodes 42338 nps 282000 hashfull 0 hashHitRate 29 pawnHitRate 852 pv g1f3 g8f6 e2e3 b8c6 f1d3 c6b4 d3h7 string thread 3
info depth 8 seldepth 8 score cp 31 time 250 nodes 93606 nps 374000 hashfull 0 hashHitRate 32 pawnHitRate 868 pv g1f3 g8f6 e2e3 b8c6 f1d3 d7d5 e1g1 e7e5 string thread 3
info depth 9 seldepth 9 score cp 31 time 510 nodes 227496 nps 446000 hashfull 1 hashHitRate 33 pawnHitRate 885 pv g1f3 g8f6 e2e3 e7e6 b1c3 f8d6 f1d3 e8g8 e1g1 string thread 3
info depth 10 seldepth 10 score cp 18 time 1450 nodes 779482 nps 537000 hashfull 5 hashHitRate 39 pawnHitRate 888 pv e2e4 b8c6 b1c3 g8f6 g1f3 e7e5 f1d3 f8d6 e1g1 e8g8 string thread 3
info depth 11 seldepth 5 score cp 29 time 4190 nodes 2540080 nps 606000 hashfull 18 hashHitRate 49 pawnHitRate 891 pv g1f3 g8f6 e2e3 d7d5 b1c3 string thread 3
info depth 12 seldepth 12 score cp 18 time 5710 nodes 3491362 nps 611000 hashfull 24 hashHitRate 50 pawnHitRate 889 pv g1f3 g8f6 e2e3 d7d5 b1c3 e7e6 d2d4 f8d6 f1d3 e8g8 e1g1 b8c6 string thread 3
info depth 13 seldepth 13 score cp 14 time 8710 nodes 5379396 nps 617000 hashfull 37 hashHitRate 51 pawnHitRate 888 pv g1f3 g8f6 e2e3 d7d5 f1e2 e7e6 c2c4 f8e7 c4d5 e6d5 e1g1 e8g8 d2d4 string thread 3
info depth 14 seldepth 14 score cp 18 time 13220 nodes 8406033 nps 635000 hashfull 50 hashHitRate 49 pawnHitRate 884 pv g1f3 g8f6 e2e3 d7d5 b1c3 e7e6 d2d4 b8c6 f1d3 f8d6 e1g1 e8g8 c1d2 c8d7 string thread 3
info depth 15 seldepth 11 score cp 19 time 21740 nodes 14127398 nps 649000 hashfull 81 hashHitRate 48 pawnHitRate 882 pv g1f3 g8f6 e2e3 d7d5 d2d4 e7e6 f1d3 f8d6 e1g1 b8c6 b1c3 string thread 3
info depth 16 seldepth 16 score cp 22 time 35640 nodes 23440512 nps 657000 hashfull 116 hashHitRate 49 pawnHitRate 882 pv g1f3 g8f6 e2e3 d7d5 d2d4 e7e6 f1d3 f8d6 e1g1 e8g8 c2c4 c7c5 c4d5 e6d5 b1c3 b8c6 string thread 3
info depth 17 seldepth 17 score cp 23 time 70080 nodes 46454370 nps 662000 hashfull 197 hashHitRate 44 pawnHitRate 886 pv g1f3 g8f6 e2e3 d7d5 d2d4 e7e6 f1d3 f8d6 e1g1 e8g8 c2c4 d5c4 d3c4 b8c6 b1c3 c8d7 c1d2 string thread
 3

What is the meaning of 'string thread 3' at the end of the PVs (all runs with one thread)?
Is there already a GUI which can show pawnHitRate (WB filtered it out)?

[Kieren Pearson]

Yes, you're correct, I have not implemented go time x or go depth x, only go infinite. Sorry about that. I'll add it to my to-do list.

Some of the output is for my own debugging. When trying to get SMP to work, I outputted which thread finished which depth at the end of the PV. In single threaded mode, you'll notice its always the same thread. The exact number is obtained using `std::this_thread::get_id();` and is meaningless. Its only so I could verify that the SMP was working and the threads were skipping depths correctly. As for hashHitRate and pawnHitRate, I only included those for my own benefit in order to gague the usefulness of each hash table. Its not part of the uci protocol and no GUI (that I know of) will actually report that statistic. As far as I know, this shouldn't be an issue and the gui should simply ignore these.

As long as this doesn't cause any issues with GUI's I'll fix it in a future release. If not, I will make a hotfix now.

[Guenther]

Yes, you're correct, I have not implemented go time x or go depth x, only go infinite. Sorry about that. I'll add it to my to-do list.

Some of the output is for my own debugging. When trying to get SMP to work, I outputted which thread finished which depth at the end of the PV. In single threaded mode, you'll notice its always the same thread. The exact number is obtained using `std::this_thread::get_id();` and is meaningless. Its only so I could verify that the SMP was working and the threads were skipping depths correctly. As for hashHitRate and pawnHitRate, I only included those for my own benefit in order to gague the usefulness of each hash table. Its not part of the uci protocol and no GUI (that I know of) will actually report that statistic. As far as I know, this shouldn't be an issue and the gui should simply ignore these.

Thanks for the infos. I just want to add a measurement for comparison for you to judge the speed of your compile.
SF 11 or SF dev versions before NNUE was intergrated are around 6-8% slower in nps on this old core2 quad.

[supersharp77]

SF NNUE 8-11-20.......

https://github.com/nodchip/Stockfish/releases

nodchip released this Aug 11, 2020 · 6 commits to master since this release

Changes:

Merged the official Stockfish master branch. (27b593a)
convert_bin Clamped score between -32,000 to 32,000. Thank you, tttak!
convert_bin Excluded invalid fens. Thank you, tttak!
convert_bin Excluded invlaid moves. Thank you, tttak!
gensfen Fixed a bug that the random seed is always constant. (#68)
gensfen Fixed a bug that the training data generator does not use EGTB. Thank you, linox! (#67)
gensfen Fixed a bug that the training data generator crashes when eval_limit is high. (#77)
gensfen detect_draw_by_consecutive_low_score option is added. If detect_draw_by_consecutive_low_score is set to 1, the training data generator adjudicates game to a draw if the last 4 moves of each engine has a score of zero starting at ply 80. The default value is 0. Thank you, linox.
gensfen detect_draw_by_insufficient_mating_material option is added. If detect_draw_by_insufficient_mating_material is set to 1, the training data generator terminates the game early and not wait for 2-fold repetition in KvK, KBvK/KvKB or KNvK/KvKN and KBvKB (same bishop color). The default value is 0. Thank you, linox.
gensfen Renamed use_draw_in_training_data_generation option to write_out_draw_game_in_training_data_generation. The old option name is still available.
learn Renamed use_draw_in_training to use_draw_games_in_training. The old option name is still available.
learn Renamed use_draw_in_validation to use_draw_games_in_validation. The old option name is still available.
learn Renamed use_hash_in_training to skip_duplicated_positions_in_training. The old option name is still available.
The network parameter file is not included in this release. Please download the strongest network parameter file on fishtest (https://tests.stockfishchess.org/nns). If there are questions, ask in the Stockfish unofficial Discord server (https://discord.gg/ZzJwPv3) at first.




https://github.com/nodchip/Stockfish/re ... 0-08-11.7z

[Damir]

The engines you just posted are for training... not playing...

[Graham Banks]

Yes, you're correct, I have not implemented go time x or go depth x, only go infinite. Sorry about that. I'll add it to my to-do list.

Some of the output is for my own debugging. When trying to get SMP to work, I outputted which thread finished which depth at the end of the PV. In single threaded mode, you'll notice its always the same thread. The exact number is obtained using `std::this_thread::get_id();` and is meaningless. Its only so I could verify that the SMP was working and the threads were skipping depths correctly. As for hashHitRate and pawnHitRate, I only included those for my own benefit in order to gague the usefulness of each hash table. Its not part of the uci protocol and no GUI (that I know of) will actually report that statistic. As far as I know, this shouldn't be an issue and the gui should simply ignore these.

As long as this doesn't cause any issues with GUI's I'll fix it in a future release. If not, I will make a hotfix now.

I cannot get either exe working, even after adding the dll files they wanted.

[Guenther]

*Edit:* (ofc I wanted to write this, just noticed now after rereading the thread)

Halogen 6 released

https://github.com/KierenP/Halogen/releases/tag/v6

In single threaded, probably about +120 ELO (2463 CCRL Blitz)
In multi threaded, probably about +180 ELO (2587 CCRL Blitz)

Both popcnt and non-popcnt builts supplied. Can anyone who uses non-popcnt builds verify it works on their machine? I have no way of making sure

Thanks for the new release. I tested the *no popcount* compile and noticed it is not a static one.
Thus people would need to dig out the missing and fitting libraries.
Better create a static one for those, who cant compile and still use old hardware.

This had now been fixed. I actually exposed a gcc bug with gcc version 8.1.0 but found this workaround.

https://stackoverflow.com/questions/100 ... lto-option

Please let me know if there are any issues.

[Gabor Szots]

I cannot get either exe working, even after adding the dll files they wanted.

Hi Graham,

I have put an exe here, without any optimizations. Maybe it works for you.