Comparison assistance on recent PEXT engine

Discussion of chess software programming and technical issues.

Moderator: Ras

Chessnut1071
Posts: 313
Joined: Tue Aug 03, 2021 2:41 pm
Full name: Bill Beame

Comparison assistance on recent PEXT engine

Post by Chessnut1071 »

First, thanks to all the great suggestions and comments from members of this board after I joined last August. At that time I had a VB engine clocking 3,890 legal moves per second. I just finished a PEXT, C# Net 6.0, bitboard version that clocks 205,000 legal moves per second on my i7 2.6 GHz laptop. My problem is I have no idea how fast it compare with engines like Herr Muhler, et. el., and those reaching speeds in the millions. I need comparisons on chips similar to my 6-year old i7, 2.6 GHz chip to see where I am. Below, is a forced checkmate that my PEXT program finds in 1,750 seconds, or, 29:10 minutes.

FEN[1201] = "5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1 ";

AI used: alpha/beta/ Zobirst Hash/ eval: check, disc check, double check, pawn promotion, ent passant, history, capture and piece type moved.

Since I'm focused on checkmate solutions the standard performance tests are inappropriate for this type of objective.

If you have an engine that solves the above FEN, please list the chip statistics for comparative purposes.
Top
User avatar
hgm
Posts: 28353
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: Comparison assistance on recent PEXT engine

Post by hgm »

Fruit 2.1 on a 3.2GHz i7 finds mate-in-7 in 1.5 sec, after searching through 3.78M positions:

Code: Select all

exclude: none best +tail                                          
dep	score	nodes	time	(not shown:  tbhits	knps	seldep)
 12	+99.87 	51.2M  	0:11.28	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
 11	+99.87 	24.8M  	0:06.35	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
 10	+99.87 	10.7M  	0:03.23	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
  9	+99.87 	5.40M  	0:01.96	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
  8	+99.87 	3.78M  	0:01.50	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
  8	+19.03 	3.36M  	0:01.36	Rd8+ Ke7 Rc8 Bf2 Qe2+ Kd6 Rd8+ Rd7 Qd2+ Kc5 Qxf2+ Kb5 Rxd7 c5 Qe2+ c4 Rxh7
  7	+17.66 	715477	0:00.31	Rd8+ Ke7 Rc8 a5 Rxc7+ Kd6 Rxf7 Bf2 Rxf6+ Kc7 Rf7+ Kd8 Qd3+ Kc8 Rxh7
  6	+15.38 	172073	0:00.09	Rd8+ Ke7 Rc8 Rf8 Qe4+ Kd7 Rxf8 Bxh6 Rf7+ Kd6 Qd4+ Ke6 Qxf6+ Kd5 Rxh7
  5	+14.34 	60660  	0:00.05	Rd8+ Ke7 Rc8 Rf8 Qe4+ Kd7 Rxf8 Bxh6 Qd4+ Ke6 Rxf6+ Ke7
  4	+13.38 	29854  	0:00.03	Rd8+ Ke7 Rc8 Rf8 Qe4+ Kf7 Rxc7+ Kg8 Qe6+ Kh8 Qxe3
  3	+11.72 	8739    0:00.02	Rd8+ Ke7 Qd3 f5 Rd7+ Ke6 Rxf7 Kxf7 Qxe3
  2	+11.72 	2549    0:00.02	Rd8+ Ke7 Qd3 f5 Rd7+ Ke6 Rxf7 Kxf7 Qxe3
  2	+10.57 	1295    0:00.00	Qb4+ c5 Qxb7 Bxh6 Qxa7
  1	+10.17 	376     0:00.00	Qb4+ Re7 Rd8+ Kf7 Qxb7
Fairy-Max takes 145 sec (221M positions) to find the mate-in-7, although it already finds a mate-in-9 after 52 sec:

Code: Select all

	mover viewpoint		fewer / Multi-PV margin = 0 / more
exclude: none best +tail                                          
dep	score	nodes	time	(not shown:  tbhits	knps	seldep)
 18	    #7 	254.1M	2:44.11	c4e4 e3d2 d1d2 f7e7 d2d8 f8f7 e4c4 e7e6 d8d7 f7e8 c4e6 e8f8 d7d8
 17	    #7 	221.5M	2:25.01	c4e4 e3h6 d1d8 f8g7 f3d4 h6g5 d4e6 g7h6 e4g4 a7a5 g4h3 g5h4 h3h4
 17	    #8 	156.3M	1:43.07	f3g5 e3g5 c4b4 c6c5 b4c5 f7e7 d1d8 f8f7 c5c4 e7e6 d8d7 f7f8 c4e6 g5h6 e6f7
 16	    #8 	110.1M	1:13.55	f3g5 e3g5 c4b4 f8e8 d1e1 e8d8 b4e4 f7f8 e4e7 d8c8 e7f8 c8d7 f8e7 d7c8 e7e8
 15	    #8 	101.7M	1:08.28	f3g5 e3g5 c4b4 f8e8 d1e1 e8d8 b4e4 f7f8 e4e7 d8c8 e7f8 c8d7 f8e7 d7c8 e7e8
 15	    #9 	76.8M  	0:51.79	d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 a5a4 f7e7 c5c4 e7e3 c4d5 c3c4 d5c4 e3d4
 14	+19.15 	47.0M  	0:31.99	d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 c5d6 f7f6 d6c5 f6e5 c5c4 e5e3 c4d5
 13	+18.04 	25.2M  	0:17.34	d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 a5a4 f7e7 c5c4 e7e3 c4d5
 12	+15.37 	9.32M  	0:06.42	d1d8 f8e7 d8c8 e3f2 c8c7 e7d6 c7f7 a7a5 f7f6 d6c7 c4f7 c7b6 f7h7 g6g5 h7e4 f2c5 f3g5
 11	+14.65 	3.52M  	0:02.43	d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 f7f6 d6c7 c4f7 c7b6 f7h7 a7a5 f6g6
 10	+12.79 	1.15M  	0:00.81	d1d8 f8e7 d8d7 e7d7 c4f7 d7d8 f7f6 d8c8 f6e6 c8d8 e6e3 c6c5 e3g5 d8c8 g5c5 a7a6
  9	+12.35 	429977	0:00.32	d1d8 f8e7 d8d7 e7d7 c4f7 d7d8 f7f6 d8c8 f6e6 c8d8
  8	+11.86 	280961	0:00.21	d1d8 f8e7 d8d7 e7d7 c4f7 d7d8 f7f6 d8e8 f6e6 e8f8 e6e3 a7a5 a2a4
  7	+10.77 	158145	0:00.14	d1d8 f8e7 c4d3 f6f5 d8d7 e7e6 d7f7 e6f7 d3e3
  6	+9.63 	27493  	0:00.03	d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6d7 f6f7 d7d6 f7h7
  5	+8.68 	12215  	0:00.01	d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6d5 f6e5
  4	+8.51 	4472    0:00.00	d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7h7
  3	+7.62 	3289    0:00.00	d1d8 f8e7 d8d7 e7d7 c4f7
  3	+7.48 	2402    0:00.00	c4b4 c6c5 b4b7 e3h6
  2	+7.48 	297     0:00.00	c4b4 f7e7 b4b7
  2	+7.17 	159     0:00.00	c4h4 a7a5
  1	+7.30 	47      0:00.00	c4h4
Both are mailbox engines.
Top
User avatar
Ras
Posts: 2696
Joined: Tue Aug 30, 2016 8:19 pm
Full name: Rasmus Althoff

Re: Comparison assistance on recent PEXT engine

Post by Ras »

Chessnut1071 wrote: ↑Thu Jun 16, 2022 6:22 amSince I'm focused on checkmate solutions
Then adding mate distance pruning could be useful.
If you have an engine that solves the above FEN, please list the chip statistics for comparative purposes.
The CT800 current development version (around 2500 Elo CCRL, single thread, mailbox) finds a mate in 9 after 3.9 seconds, mate in 8 after 7.3 seconds, mate in 7 after 11.5 seconds (24M nodes, 2.1M nodes/s).

That's on a Ryzen 7 5700G with boost disabled, i.e. @3.8GHz (costs ~ 13% single thread performance). Check out CPU Benchmark to get a rough idea about the CPU performance relative to yours, especially single thread.

Code: Select all

info depth 2 seldepth 4 score cp 931 time 22 nodes 623 nps 28318 hashfull 0 tbhits 0 pv d1d8 f8e7 d8d7 e7d7
info depth 3 seldepth 6 score cp 931 time 22 nodes 1405 nps 63863 hashfull 0 tbhits 0 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6
info depth 4 seldepth 6 score cp 957 time 23 nodes 3249 nps 141260 hashfull 0 tbhits 0 pv c4b4 c6c5 b4b7 e3h6 b7a7 h6e3
info depth 5 seldepth 8 score cp 1057 time 24 nodes 5167 nps 215291 hashfull 0 tbhits 0 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7h7 c6c5
info depth 6 seldepth 12 score cp 1183 time 25 nodes 7851 nps 314040 hashfull 0 tbhits 0 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6d7 f3e5 d7e8 f6h8 e8e7
info depth 7 seldepth 12 score cp 1183 time 26 nodes 8334 nps 320538 hashfull 0 tbhits 0 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6d7 f3e5 d7e8 f6h8 e8e7
info depth 8 seldepth 14 score cp 1226 time 28 nodes 12885 nps 460178 hashfull 0 tbhits 0 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6c5 f6e5 c5c4 e5e4 c4b5 a2a4 b5b6
info depth 9 seldepth 14 score cp 1226 time 30 nodes 16195 nps 539833 hashfull 0 tbhits 0 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6c5 f6e5 c5c4 e5e4 c4b5 a2a4 b5b6
info depth 10 seldepth 14 score cp 1242 time 36 nodes 30811 nps 855861 hashfull 0 tbhits 0 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6c5 f6e5 c5c4 e5e3 a7a5 e3f4 c4d5
info depth 11 seldepth 12 score cp 1495 time 78 nodes 112578 nps 1443307 hashfull 1 tbhits 0 pv d1d8 f8e7 d8c8 e3f4 c4e4 e7d7 e4e8 d7d6 e8f7 d6c5 f7h7 g6g5
info depth 12 seldepth 14 score cp 1586 time 349 nodes 612023 nps 1753647 hashfull 10 tbhits 0 pv d1d8 f8e7 d8c8 e3b6 c4e4 e7d7 e4e8 d7d6 e8f7 f6f5 c8d8 d6c5 f7h7 c5c4
info depth 13 seldepth 14 score cp 1586 time 380 nodes 670734 nps 1765089 hashfull 10 tbhits 0 pv d1d8 f8e7 d8c8 e3b6 c4e4 e7d7 e4e8 d7d6 e8f7 d6c5 f7h7 c5d6 c8d8 d6e6
info depth 14 seldepth 18 score cp 1788 time 904 nodes 1611793 nps 1782956 hashfull 24 tbhits 0 pv d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 c4d3 d6c5 b2b4 c5b6 d3d8 b6b5 f7b7 b5a6 d8d7 c6c5
info time 1000 nodes 1789834 nps 1789834 hashfull 27 tbhits 0 currmove c4b4 currmovenumber 2
info depth 15 seldepth 18 score cp 1819 time 1018 nodes 1824434 nps 1792174 hashfull 27 tbhits 0 pv d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 c4d3 d6c5 b2b4 c5b6 d3d8 b6b5 f7b7 b5a6 d8d7 c6c5
info depth 16 seldepth 18 score cp 2256 time 1834 nodes 3329209 nps 1815272 hashfull 52 tbhits 0 pv d1d8 f8e7 d8c8 e7d6 c4d3 d6c5 d3e3 c5b5 c3c4 b5c4 e3e6 c4c5 e6f7 g6g5 f7h7 c5c4 h7e4 c4c5
info time 2000 nodes 3629098 nps 1814549 hashfull 58 tbhits 0 currmove d1d8 currmovenumber 1
info depth 17 seldepth 18 score cp 2381 time 2387 nodes 4341762 nps 1818919 hashfull 70 tbhits 0 pv d1d8 f8e7 d8c8 e7d6 c4d3 d6c5 b2b4 c5b6 d3e3 b6a6 e3e6 b7b6 e6f7 c6c5 f7h7 c5b4 c3b4 a6b5
info time 3000 nodes 5533480 nps 1844493 hashfull 89 tbhits 0 currmove d1d8 currmovenumber 1
info depth 18 seldepth 18 score mate 9 time 3861 nodes 7155164 nps 1853189 hashfull 109 tbhits 0 pv d1d8 f8e7 d8c8 a7a5 c4e4 e7d6 e4e3 c6c5 c8e8 f7e7 e3e7 d6c6 e7e6 c6b5 a2a4 b5a4 e6c4
info time 4000 nodes 7417862 nps 1854465 hashfull 112 tbhits 0 currmove d1d8 currmovenumber 1
info depth 19 seldepth 19 score mate 9 time 4229 nodes 7891018 nps 1865930 hashfull 116 tbhits 0 pv d1d8 f8e7 d8c8 a7a5 c4e4 e7d6 e4e3 c6c5 c8e8 f7e7 e3e7 d6c6 e7e6 c6b5 a2a4 b5a4 e6c4
info depth 20 seldepth 20 score mate 9 time 4673 nodes 8753761 nps 1873263 hashfull 124 tbhits 0 pv d1d8 f8e7 d8c8 a7a5 c4e4 e7d6 e4e3 c6c5 c8e8 f7e7 e3e7 d6c6 e7e6 c6b5 a2a4 b5a4 e6c4
info time 5000 nodes 9315128 nps 1863025 hashfull 131 tbhits 0 currmove d1d8 currmovenumber 1
info depth 21 seldepth 21 score mate 9 time 5566 nodes 10473470 nps 1881687 hashfull 142 tbhits 0 pv d1d8 f8e7 d8c8 a7a5 c4e4 e7d6 e4e3 c6c5 c8e8 f7e7 e3e7 d6c6 e7e6 c6b5 a2a4 b5a4 e6c4
info time 6000 nodes 11300393 nps 1883398 hashfull 149 tbhits 0 currmove d1d8 currmovenumber 1
info time 7000 nodes 13334826 nps 1904975 hashfull 171 tbhits 0 currmove d1d3 currmovenumber 10
info depth 22 seldepth 22 score mate 8 time 7342 nodes 14064454 nps 1915616 hashfull 179 tbhits 0 pv c4e6 e3h6 d1d8 f8g7 f3e5 f6e5 e6e5 f7f6 d8d7 g7g8 e5f6 a7a5 f6f7 g8h8 f7h7
info time 8000 nodes 15654218 nps 1956777 hashfull 186 tbhits 0 currmove f3d4 currmovenumber 23
info depth 23 seldepth 23 score mate 8 time 8177 nodes 16052977 nps 1963186 hashfull 188 tbhits 0 pv c4e6 e3h6 d1d8 f8g7 f3e5 f6e5 e6e5 f7f6 d8d7 g7g8 e5f6 a7a5 f6f7 g8h8 f7h7
info time 9001 nodes 17992280 nps 1998920 hashfull 201 tbhits 0 currmove d1d3 currmovenumber 10
info depth 24 seldepth 24 score mate 8 time 9405 nodes 18916781 nps 2011353 hashfull 206 tbhits 0 pv c4e6 e3h6 d1d8 f8g7 f3e5 f6e5 e6e5 f7f6 d8d7 g7g8 e5f6 a7a5 f6f7 g8h8 f7h7
info time 10000 nodes 20524586 nps 2052458 hashfull 212 tbhits 0 currmove d1d8 currmovenumber 2
info time 11000 nodes 23012002 nps 2092000 hashfull 223 tbhits 0 currmove c4e4 currmovenumber 16
info depth 25 seldepth 25 score mate 7 time 11509 nodes 24316890 nps 2112858 hashfull 232 tbhits 0 pv c4e4 e3h6 d1d8 f8g7 f3d4 h6f4 d4e6 g7h6 d8d1 c6c5 d1h1 f4h2 h1h2
info time 12000 nodes 25782743 nps 2148561 hashfull 237 tbhits 0 currmove c4b4 currmovenumber 4
info depth 26 seldepth 26 score mate 7 time 12479 nodes 27171344 nps 2177365 hashfull 241 tbhits 0 pv c4e4 e3h6 d1d8 f8g7 f3d4 h6f4 d4e6 g7h6 d8d1 c6c5 d1h1 f4h2 h1h2
info time 13000 nodes 28706488 nps 2208191 hashfull 245 tbhits 0 currmove d1d8 currmovenumber 3
info depth 27 seldepth 27 score mate 7 time 13877 nodes 31216148 nps 2249488 hashfull 255 tbhits 0 pv c4e4 e3h6 d1d8 f8g7 f3d4 h6f4 d4e6 g7h6 d8d1 c6c5 d1h1 f4h2 h1h2
Rasmus Althoff
https://www.ct800.net
Top
User avatar
lithander
Posts: 915
Joined: Sun Dec 27, 2020 2:40 am
Location: Bremen, Germany
Full name: Thomas Jahn

Re: Comparison assistance on recent PEXT engine

Post by lithander »

Chessnut1071 wrote: ↑Thu Jun 16, 2022 6:22 am If you have an engine that solves the above FEN, please list the chip statistics for comparative purposes.
Leorik finds a mate in 8 in 1.5s and the final mate in 7 in 5.2s on a i7-9700K CPU

Code: Select all

Leorik 2.1.2
position fen 5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1
go
info depth 1 score cp 858 nodes 232 nps 116000 time 2 pv c4b4
info depth 2 score cp 858 nodes 332 nps 23714 time 14 pv c4b4 f8g8
info depth 3 score cp 897 nodes 668 nps 44533 time 15 pv d1d8 f8e7 d8d7
info depth 4 score cp 897 nodes 1070 nps 71333 time 15 pv d1d8 f8e7 d8d7 e7d7
info depth 5 score cp 897 nodes 1792 nps 112000 time 16 pv d1d8 f8e7 d8d7 e7d7 c4f7
info depth 6 score cp 897 nodes 3071 nps 191937 time 16 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6
info depth 7 score cp 897 nodes 6348 nps 352666 time 18 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7h7
info depth 8 score cp 927 nodes 17477 nps 832238 time 21 pv d1d8 f8e7 d8b8 e3b6 b8b7 c6c5 c4d5 f6f5
info depth 9 score cp 1195 nodes 49687 nps 1552718 time 32 pv d1d8 f8e7 c4d3 f6f5 f3e5 e3h6 d3d7 e7f6 e5f7
info depth 10 score cp 1213 nodes 76133 nps 1856902 time 41 pv d1d8 f8e7 d8c8 f7f8 c4e4 e7f7 c8c7 f7g8 e4e6 g8h8
info depth 11 score cp 1263 nodes 142792 nps 2196800 time 65 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d6 c7b7 c6c5 c4d3 e3d4 c3d4
info depth 12 score cp 1321 nodes 203391 nps 2311261 time 88 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d8 c7h7 b7b5 c4d3 d8c8 d3e3 a7a5
info depth 13 score cp 1628 nodes 944382 nps 2861763 time 330 pv d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c7f7 e3f2 f7b7 a5a4 b7h7 g6g5 c4a4
info depth 14 score cp 1651 nodes 1145291 nps 3054109 time 375 pv d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c7f7 e3f2 f7f6 d6c7 c4g8 c7b6 g8h7 a5a4
info depth 15 score cp 1902 nodes 2397229 nps 4140291 time 579 pv d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 a5a4 f7e7 c5c4 e7e3 c4d5 e3d4
info depth 16 score cp 1902 nodes 2978104 nps 4260520 time 699 pv d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 a5a4 f7e7 c5c4 e7e3 c4d5 e3d4 d5e6
info depth 17 score mate 8 nodes 7520552 nps 4934745 time 1524 pv c4e6 e3h6 d1d8 f8g7 f3e5 f6e5 e6e5 f7f6 d8d7 g7g8 e5f6 h6f8 f6f7 g8h8 f7f8
info depth 18 score mate 8 nodes 12203731 nps 4823609 time 2530 pv c4e6 e3h6 d1d8 f8g7 f3e5 f6e5 e6e5 f7f6 d8d7 g7g8 e5f6 h6f8 f6f7 g8h8 f7f8
info depth 19 score mate 7 nodes 26917998 nps 5191513 time 5185 pv c4e4 e3h6 d1d8 f8g7 f3d4 h6f4 d4e6 g7h6 d8d1 f7e7 d1h1 f4h2 h1h2
stop
bestmove c4e4
Minimal Chess (simple, open source, C#) - Youtube & Github
Leorik (competitive, in active development, C#) - Github & Lichess
Top
Mergi
Posts: 127
Joined: Sat Aug 21, 2021 9:55 pm
Full name: Jen

Re: Comparison assistance on recent PEXT engine

Post by Mergi »

Wouldn't it make more sense to download other engines for yourself so you are able to measure and test on the same hardware? The CCRL list is a great resource for that, as most engines also have a link to the executable/source code there. Comparisons like these seem a bit pointless.
Top
Chessnut1071
Posts: 313
Joined: Tue Aug 03, 2021 2:41 pm
Full name: Bill Beame

Re: Comparison assistance on recent PEXT engine

Post by Chessnut1071 »

Chessnut1071 wrote: ↑Thu Jun 16, 2022 6:22 am First, thanks to all the great suggestions and comments from members of this board after I joined last August. At that time I had a VB engine clocking 3,890 legal moves per second. I just finished a PEXT, C# Net 6.0, bitboard version that clocks 205,000 legal moves per second on my i7 2.6 GHz laptop. My problem is I have no idea how fast it compare with engines like Herr Muhler, et. el., and those reaching speeds in the millions. I need comparisons on chips similar to my 6-year old i7, 2.6 GHz chip to see where I am. Below, is a forced checkmate that my PEXT program finds in 1,750 seconds, or, 29:10 minutes.

FEN[1201] = "5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1 ";

AI used: alpha/beta/ Zobirst Hash/ eval: check, disc check, double check, pawn promotion, ent passant, history, capture and piece type moved.

Since I'm focused on checkmate solutions the standard performance tests are inappropriate for this type of objective.

If you have an engine that solves the above FEN, please list the chip statistics for comparative purposes.
Positions examined: legal moves [331,196,982] for 189,255 legal moves/sec.
Top
Chessnut1071
Posts: 313
Joined: Tue Aug 03, 2021 2:41 pm
Full name: Bill Beame

Re: Comparison assistance on recent PEXT engine

Post by Chessnut1071 »

hgm wrote: ↑Thu Jun 16, 2022 9:18 am Fruit 2.1 on a 3.2GHz i7 finds mate-in-7 in 1.5 sec, after searching through 3.78M positions:

Code: Select all

exclude: none best +tail                                          
dep	score	nodes	time	(not shown:  tbhits	knps	seldep)
 12	+99.87 	51.2M  	0:11.28	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
 11	+99.87 	24.8M  	0:06.35	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
 10	+99.87 	10.7M  	0:03.23	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
  9	+99.87 	5.40M  	0:01.96	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
  8	+99.87 	3.78M  	0:01.50	Qe4 Bxh6 Rd8+ Kg7 Nd4 Rf8 Qe7+ Kg8 Rd7 Rf7 Qxf7+ Kh8 Qxh7#
  8	+19.03 	3.36M  	0:01.36	Rd8+ Ke7 Rc8 Bf2 Qe2+ Kd6 Rd8+ Rd7 Qd2+ Kc5 Qxf2+ Kb5 Rxd7 c5 Qe2+ c4 Rxh7
  7	+17.66 	715477	0:00.31	Rd8+ Ke7 Rc8 a5 Rxc7+ Kd6 Rxf7 Bf2 Rxf6+ Kc7 Rf7+ Kd8 Qd3+ Kc8 Rxh7
  6	+15.38 	172073	0:00.09	Rd8+ Ke7 Rc8 Rf8 Qe4+ Kd7 Rxf8 Bxh6 Rf7+ Kd6 Qd4+ Ke6 Qxf6+ Kd5 Rxh7
  5	+14.34 	60660  	0:00.05	Rd8+ Ke7 Rc8 Rf8 Qe4+ Kd7 Rxf8 Bxh6 Qd4+ Ke6 Rxf6+ Ke7
  4	+13.38 	29854  	0:00.03	Rd8+ Ke7 Rc8 Rf8 Qe4+ Kf7 Rxc7+ Kg8 Qe6+ Kh8 Qxe3
  3	+11.72 	8739    0:00.02	Rd8+ Ke7 Qd3 f5 Rd7+ Ke6 Rxf7 Kxf7 Qxe3
  2	+11.72 	2549    0:00.02	Rd8+ Ke7 Qd3 f5 Rd7+ Ke6 Rxf7 Kxf7 Qxe3
  2	+10.57 	1295    0:00.00	Qb4+ c5 Qxb7 Bxh6 Qxa7
  1	+10.17 	376     0:00.00	Qb4+ Re7 Rd8+ Kf7 Qxb7
Fairy-Max takes 145 sec (221M positions) to find the mate-in-7, although it already finds a mate-in-9 after 52 sec:

Code: Select all

	mover viewpoint		fewer / Multi-PV margin = 0 / more
exclude: none best +tail                                          
dep	score	nodes	time	(not shown:  tbhits	knps	seldep)
 18	    #7 	254.1M	2:44.11	c4e4 e3d2 d1d2 f7e7 d2d8 f8f7 e4c4 e7e6 d8d7 f7e8 c4e6 e8f8 d7d8
 17	    #7 	221.5M	2:25.01	c4e4 e3h6 d1d8 f8g7 f3d4 h6g5 d4e6 g7h6 e4g4 a7a5 g4h3 g5h4 h3h4
 17	    #8 	156.3M	1:43.07	f3g5 e3g5 c4b4 c6c5 b4c5 f7e7 d1d8 f8f7 c5c4 e7e6 d8d7 f7f8 c4e6 g5h6 e6f7
 16	    #8 	110.1M	1:13.55	f3g5 e3g5 c4b4 f8e8 d1e1 e8d8 b4e4 f7f8 e4e7 d8c8 e7f8 c8d7 f8e7 d7c8 e7e8
 15	    #8 	101.7M	1:08.28	f3g5 e3g5 c4b4 f8e8 d1e1 e8d8 b4e4 f7f8 e4e7 d8c8 e7f8 c8d7 f8e7 d7c8 e7e8
 15	    #9 	76.8M  	0:51.79	d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 a5a4 f7e7 c5c4 e7e3 c4d5 c3c4 d5c4 e3d4
 14	+19.15 	47.0M  	0:31.99	d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 c5d6 f7f6 d6c5 f6e5 c5c4 e5e3 c4d5
 13	+18.04 	25.2M  	0:17.34	d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 a5a4 f7e7 c5c4 e7e3 c4d5
 12	+15.37 	9.32M  	0:06.42	d1d8 f8e7 d8c8 e3f2 c8c7 e7d6 c7f7 a7a5 f7f6 d6c7 c4f7 c7b6 f7h7 g6g5 h7e4 f2c5 f3g5
 11	+14.65 	3.52M  	0:02.43	d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 f7f6 d6c7 c4f7 c7b6 f7h7 a7a5 f6g6
 10	+12.79 	1.15M  	0:00.81	d1d8 f8e7 d8d7 e7d7 c4f7 d7d8 f7f6 d8c8 f6e6 c8d8 e6e3 c6c5 e3g5 d8c8 g5c5 a7a6
  9	+12.35 	429977	0:00.32	d1d8 f8e7 d8d7 e7d7 c4f7 d7d8 f7f6 d8c8 f6e6 c8d8
  8	+11.86 	280961	0:00.21	d1d8 f8e7 d8d7 e7d7 c4f7 d7d8 f7f6 d8e8 f6e6 e8f8 e6e3 a7a5 a2a4
  7	+10.77 	158145	0:00.14	d1d8 f8e7 c4d3 f6f5 d8d7 e7e6 d7f7 e6f7 d3e3
  6	+9.63 	27493  	0:00.03	d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6d7 f6f7 d7d6 f7h7
  5	+8.68 	12215  	0:00.01	d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7f6 d6d5 f6e5
  4	+8.51 	4472    0:00.00	d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7h7
  3	+7.62 	3289    0:00.00	d1d8 f8e7 d8d7 e7d7 c4f7
  3	+7.48 	2402    0:00.00	c4b4 c6c5 b4b7 e3h6
  2	+7.48 	297     0:00.00	c4b4 f7e7 b4b7
  2	+7.17 	159     0:00.00	c4h4 a7a5
  1	+7.30 	47      0:00.00	c4h4
Both are mailbox engines.
Fruit 2.1 on a 3.2GHz i7 finds mate-in-7 in 1.5 sec, after searching through 3.78M positions:

Fairy-Max takes 145 sec (221M positions) to find the mate-in-7, although it already finds a mate-in-9 after 52 sec:

I used 331,196,982 legal calls for 1750 seconds, similar to Fairy_max, but, a lot slower, about 1/10 the speed. Engines that find that mate in less than 30 million searches must be using a very advanced evaluation. My evaluation consists of:
Priority:
1) double check, disc check, check
2) pawn promotion
3) capture (type)
4) history
5) ent passant

This is all I can pack into 12 bits in my 64-bit move variable. The other 52 bits are data associated with the move:
1) piece 2) move 3) capture 4) direction 5) check type 6) check: from 7) check: to 8) check direction 9) skip bit 10) pawn promotion 11) e.p. bit

If Fruit can solve that mate after searching only 3.78 positions it must be using something more sophisticated than my simple eval metrics above.

?, are mailbox engines faster than PEXT and magic boards on mate solutions? All that data slows down my fast PEXT engine speed, but, without the data there's no evaluation. The evaluation seems to be the key to reaching those incredible speeds reported here.
Top
Chessnut1071
Posts: 313
Joined: Tue Aug 03, 2021 2:41 pm
Full name: Bill Beame

Re: Comparison assistance on recent PEXT engine

Post by Chessnut1071 »

lithander wrote: ↑Thu Jun 16, 2022 11:23 am
Chessnut1071 wrote: ↑Thu Jun 16, 2022 6:22 am If you have an engine that solves the above FEN, please list the chip statistics for comparative purposes.
Leorik finds a mate in 8 in 1.5s and the final mate in 7 in 5.2s on a i7-9700K CPU

Code: Select all

Leorik 2.1.2
position fen 5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1
go
info depth 1 score cp 858 nodes 232 nps 116000 time 2 pv c4b4
info depth 2 score cp 858 nodes 332 nps 23714 time 14 pv c4b4 f8g8
info depth 3 score cp 897 nodes 668 nps 44533 time 15 pv d1d8 f8e7 d8d7
info depth 4 score cp 897 nodes 1070 nps 71333 time 15 pv d1d8 f8e7 d8d7 e7d7
info depth 5 score cp 897 nodes 1792 nps 112000 time 16 pv d1d8 f8e7 d8d7 e7d7 c4f7
info depth 6 score cp 897 nodes 3071 nps 191937 time 16 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6
info depth 7 score cp 897 nodes 6348 nps 352666 time 18 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7d6 f7h7
info depth 8 score cp 927 nodes 17477 nps 832238 time 21 pv d1d8 f8e7 d8b8 e3b6 b8b7 c6c5 c4d5 f6f5
info depth 9 score cp 1195 nodes 49687 nps 1552718 time 32 pv d1d8 f8e7 c4d3 f6f5 f3e5 e3h6 d3d7 e7f6 e5f7
info depth 10 score cp 1213 nodes 76133 nps 1856902 time 41 pv d1d8 f8e7 d8c8 f7f8 c4e4 e7f7 c8c7 f7g8 e4e6 g8h8
info depth 11 score cp 1263 nodes 142792 nps 2196800 time 65 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d6 c7b7 c6c5 c4d3 e3d4 c3d4
info depth 12 score cp 1321 nodes 203391 nps 2311261 time 88 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d8 c7h7 b7b5 c4d3 d8c8 d3e3 a7a5
info depth 13 score cp 1628 nodes 944382 nps 2861763 time 330 pv d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c7f7 e3f2 f7b7 a5a4 b7h7 g6g5 c4a4
info depth 14 score cp 1651 nodes 1145291 nps 3054109 time 375 pv d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c7f7 e3f2 f7f6 d6c7 c4g8 c7b6 g8h7 a5a4
info depth 15 score cp 1902 nodes 2397229 nps 4140291 time 579 pv d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 a5a4 f7e7 c5c4 e7e3 c4d5 e3d4
info depth 16 score cp 1902 nodes 2978104 nps 4260520 time 699 pv d1d8 f8e7 d8c8 a7a5 c8c7 e7d6 c4f7 d6c5 c7b7 a5a4 f7e7 c5c4 e7e3 c4d5 e3d4 d5e6
info depth 17 score mate 8 nodes 7520552 nps 4934745 time 1524 pv c4e6 e3h6 d1d8 f8g7 f3e5 f6e5 e6e5 f7f6 d8d7 g7g8 e5f6 h6f8 f6f7 g8h8 f7f8
info depth 18 score mate 8 nodes 12203731 nps 4823609 time 2530 pv c4e6 e3h6 d1d8 f8g7 f3e5 f6e5 e6e5 f7f6 d8d7 g7g8 e5f6 h6f8 f6f7 g8h8 f7f8
info depth 19 score mate 7 nodes 26917998 nps 5191513 time 5185 pv c4e4 e3h6 d1d8 f8g7 f3d4 h6f4 d4e6 g7h6 d8d1 f7e7 d1h1 f4h2 h1h2
stop
bestmove c4e4
? any info on what's in that evaluation. I had to examine over 10x the positions. Up until today, I thought it was pretty efficient. weeping, weeping.
Top
Chessnut1071
Posts: 313
Joined: Tue Aug 03, 2021 2:41 pm
Full name: Bill Beame

Re: Comparison assistance on recent PEXT engine

Post by Chessnut1071 »

Mergi wrote: ↑Thu Jun 16, 2022 11:31 am Wouldn't it make more sense to download other engines for yourself so you are able to measure and test on the same hardware? The CCRL list is a great resource for that, as most engines also have a link to the executable/source code there. Comparisons like these seem a bit pointless.
thx for the suggestion. That's probably my next step.
Top
User avatar
lithander
Posts: 915
Joined: Sun Dec 27, 2020 2:40 am
Location: Bremen, Germany
Full name: Thomas Jahn

Re: Comparison assistance on recent PEXT engine

Post by lithander »

Chessnut1071 wrote: ↑Thu Jun 16, 2022 1:31 pm ? any info on what's in that evaluation. I had to examine over 10x the positions. Up until today, I thought it was pretty efficient. weeping, weeping.
I think for finding these mates the search is more important than the evaluation. Version 2.0 of Leorik is equally fast and in that version the eval is still just tapered PSQTs.
Minimal Chess (simple, open source, C#) - Youtube & Github
Leorik (competitive, in active development, C#) - Github & Lichess
Top

Return to “Computer Chess Club: Programming and Technical Discussions”