Comparison wanted: mate in 7 test

[tcusr]

I'm testing my new engine which is based on mailbox ideas; however, it's not really a mailbox in a formal sense. It's 21x faster than my old inefficient engine, but, I don't have any idea how it compares with a well structured mailbox or bitboard program. So, I tested it on the FEN below with the following statistics:

FEN[201] = "5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1 "; // Unknown author 7-move mate [Qe4]

engine calls: 847,550,776
nodes visited: 34,141233,773
time: 11,327.6 seconds

methods: alpha/beta, history by ply, Zobirst hash, optimized evaluation function: history, capture, check, slider piece, enemy king moves[feedback[], and pawn promotion.

computer chip Intel I5, 2.5 GHz, 16GB ram

Need somebody to compare with.

My engine doesn't have history heuristics yet, nor any prunings except alpha/beta itself. It finds mate in 7 at depth 10:

Code: Select all

go
info score cp 822 depth 1 seldepth 6 time 0 nodes 274 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 822 depth 2 seldepth 7 time 0 nodes 886 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 872 depth 3 seldepth 10 time 3 nodes 4935 nps 1645000 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7c8 f7h7
info score cp 1098 depth 4 seldepth 11 time 7 nodes 23304 nps 3329143 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1098 depth 5 seldepth 13 time 20 nodes 119780 nps 5989000 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1170 depth 6 seldepth 15 time 93 nodes 650253 nps 6991968 hashfull 1 pv c4e4 e3h6 d1d8 f8g7 e4e8 h6f4 e8g8 g7h6 g8f7
info score cp 1264 depth 7 seldepth 17 time 403 nodes 2871969 nps 7126474 hashfull 5 pv d1d8 f8e7 d8c8 f7f8 c4e4 e7f7 c8c7 f7g8 e4e3 f8f7 c7f7 g8f7 e3a7
info score cp 1312 depth 8 seldepth 21 time 1696 nodes 12136697 nps 7156071 hashfull 22 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d6 c7b7 c6c5 c4d3 e3d4 c3d4
info score cp 1549 depth 9 seldepth 21 time 7216 nodes 50730558 nps 7030288 hashfull 95 pv d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 f7b7 a7a5 b7h7
info score mate 7 depth 10 seldepth 23 time 38200 nodes 270719243 nps 7086891 hashfull 399 pv c4e4 e3h6 d1d8 f8g7 f3d4 f7f8 e4e7 g7g8 d8d7 f8f7 e7f7 g8h8 f7h7

38 seconds on a Core i7-6700K.

that's really impressive!
you must have a really good evaluation function then, i only have mvv/lva ordering and delta pruning but after 10 minutes it still hasn't finished
(your psq tables don't seem computer generated, did you write them yourself?)

[Chessnut1071]

I'm testing my new engine which is based on mailbox ideas; however, it's not really a mailbox in a formal sense. It's 21x faster than my old inefficient engine, but, I don't have any idea how it compares with a well structured mailbox or bitboard program. So, I tested it on the FEN below with the following statistics:

FEN[201] = "5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1 "; // Unknown author 7-move mate [Qe4]

engine calls: 847,550,776
nodes visited: 34,141233,773
time: 11,327.6 seconds

methods: alpha/beta, history by ply, Zobirst hash, optimized evaluation function: history, capture, check, slider piece, enemy king moves[feedback[], and pawn promotion.

computer chip Intel I5, 2.5 GHz, 16GB ram

Need somebody to compare with.

My engine doesn't have history heuristics yet, nor any prunings except alpha/beta itself. It finds mate in 7 at depth 10:

Code: Select all

go
info score cp 822 depth 1 seldepth 6 time 0 nodes 274 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 822 depth 2 seldepth 7 time 0 nodes 886 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 872 depth 3 seldepth 10 time 3 nodes 4935 nps 1645000 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7c8 f7h7
info score cp 1098 depth 4 seldepth 11 time 7 nodes 23304 nps 3329143 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1098 depth 5 seldepth 13 time 20 nodes 119780 nps 5989000 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1170 depth 6 seldepth 15 time 93 nodes 650253 nps 6991968 hashfull 1 pv c4e4 e3h6 d1d8 f8g7 e4e8 h6f4 e8g8 g7h6 g8f7
info score cp 1264 depth 7 seldepth 17 time 403 nodes 2871969 nps 7126474 hashfull 5 pv d1d8 f8e7 d8c8 f7f8 c4e4 e7f7 c8c7 f7g8 e4e3 f8f7 c7f7 g8f7 e3a7
info score cp 1312 depth 8 seldepth 21 time 1696 nodes 12136697 nps 7156071 hashfull 22 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d6 c7b7 c6c5 c4d3 e3d4 c3d4
info score cp 1549 depth 9 seldepth 21 time 7216 nodes 50730558 nps 7030288 hashfull 95 pv d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 f7b7 a7a5 b7h7
info score mate 7 depth 10 seldepth 23 time 38200 nodes 270719243 nps 7086891 hashfull 399 pv c4e4 e3h6 d1d8 f8g7 f3d4 f7f8 e4e7 g7g8 d8d7 f8f7 e7f7 g8h8 f7h7

38 seconds on a Core i7-6700K.

that's really impressive!
you must have a really good evaluation function then, i only have mvv/lva ordering and delta pruning but after 10 minutes it still hasn't finished
(your psq tables don't seem computer generated, did you write them yourself?)

I don't think it's his evaluation, he's clocking 270,000,000 nodes per second. Wondering what method he's using, mailboxes or bitboards?

[Chessnut1071]

I'm testing my new engine which is based on mailbox ideas; however, it's not really a mailbox in a formal sense. It's 21x faster than my old inefficient engine, but, I don't have any idea how it compares with a well structured mailbox or bitboard program. So, I tested it on the FEN below with the following statistics:

FEN[201] = "5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1 "; // Unknown author 7-move mate [Qe4]

engine calls: 847,550,776
nodes visited: 34,141233,773
time: 11,327.6 seconds

methods: alpha/beta, history by ply, Zobirst hash, optimized evaluation function: history, capture, check, slider piece, enemy king moves[feedback[], and pawn promotion.

computer chip Intel I5, 2.5 GHz, 16GB ram

Need somebody to compare with.

My engine doesn't have history heuristics yet, nor any prunings except alpha/beta itself. It finds mate in 7 at depth 10:

Code: Select all

go
info score cp 822 depth 1 seldepth 6 time 0 nodes 274 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 822 depth 2 seldepth 7 time 0 nodes 886 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 872 depth 3 seldepth 10 time 3 nodes 4935 nps 1645000 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7c8 f7h7
info score cp 1098 depth 4 seldepth 11 time 7 nodes 23304 nps 3329143 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1098 depth 5 seldepth 13 time 20 nodes 119780 nps 5989000 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1170 depth 6 seldepth 15 time 93 nodes 650253 nps 6991968 hashfull 1 pv c4e4 e3h6 d1d8 f8g7 e4e8 h6f4 e8g8 g7h6 g8f7
info score cp 1264 depth 7 seldepth 17 time 403 nodes 2871969 nps 7126474 hashfull 5 pv d1d8 f8e7 d8c8 f7f8 c4e4 e7f7 c8c7 f7g8 e4e3 f8f7 c7f7 g8f7 e3a7
info score cp 1312 depth 8 seldepth 21 time 1696 nodes 12136697 nps 7156071 hashfull 22 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d6 c7b7 c6c5 c4d3 e3d4 c3d4
info score cp 1549 depth 9 seldepth 21 time 7216 nodes 50730558 nps 7030288 hashfull 95 pv d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 f7b7 a7a5 b7h7
info score mate 7 depth 10 seldepth 23 time 38200 nodes 270719243 nps 7086891 hashfull 399 pv c4e4 e3h6 d1d8 f8g7 f3d4 f7f8 e4e7 g7g8 d8d7 f8f7 e7f7 g8h8 f7h7

38 seconds on a Core i7-6700K.

that's really impressive!
you must have a really good evaluation function then, i only have mvv/lva ordering and delta pruning but after 10 minutes it still hasn't finished
(your psq tables don't seem computer generated, did you write them yourself?)

I don't think it's his evaluation, he's clocking 270,000,000 nodes per second. Wondering what method he's using, mailboxes or bitboards?

[Chessnut1071]

I'm testing my new engine which is based on mailbox ideas; however, it's not really a mailbox in a formal sense. It's 21x faster than my old inefficient engine, but, I don't have any idea how it compares with a well structured mailbox or bitboard program. So, I tested it on the FEN below with the following statistics:

FEN[201] = "5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1 "; // Unknown author 7-move mate [Qe4]

engine calls: 847,550,776
nodes visited: 34,141233,773
time: 11,327.6 seconds

methods: alpha/beta, history by ply, Zobirst hash, optimized evaluation function: history, capture, check, slider piece, enemy king moves[feedback[], and pawn promotion.

computer chip Intel I5, 2.5 GHz, 16GB ram

Need somebody to compare with.

My engine doesn't have history heuristics yet, nor any prunings except alpha/beta itself. It finds mate in 7 at depth 10:

Code: Select all

go
info score cp 822 depth 1 seldepth 6 time 0 nodes 274 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 822 depth 2 seldepth 7 time 0 nodes 886 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 872 depth 3 seldepth 10 time 3 nodes 4935 nps 1645000 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7c8 f7h7
info score cp 1098 depth 4 seldepth 11 time 7 nodes 23304 nps 3329143 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1098 depth 5 seldepth 13 time 20 nodes 119780 nps 5989000 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1170 depth 6 seldepth 15 time 93 nodes 650253 nps 6991968 hashfull 1 pv c4e4 e3h6 d1d8 f8g7 e4e8 h6f4 e8g8 g7h6 g8f7
info score cp 1264 depth 7 seldepth 17 time 403 nodes 2871969 nps 7126474 hashfull 5 pv d1d8 f8e7 d8c8 f7f8 c4e4 e7f7 c8c7 f7g8 e4e3 f8f7 c7f7 g8f7 e3a7
info score cp 1312 depth 8 seldepth 21 time 1696 nodes 12136697 nps 7156071 hashfull 22 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d6 c7b7 c6c5 c4d3 e3d4 c3d4
info score cp 1549 depth 9 seldepth 21 time 7216 nodes 50730558 nps 7030288 hashfull 95 pv d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 f7b7 a7a5 b7h7
info score mate 7 depth 10 seldepth 23 time 38200 nodes 270719243 nps 7086891 hashfull 399 pv c4e4 e3h6 d1d8 f8g7 f3d4 f7f8 e4e7 g7g8 d8d7 f8f7 e7f7 g8h8 f7h7

38 seconds on a Core i7-6700K.

that's really impressive!
you must have a really good evaluation function then, i only have mvv/lva ordering and delta pruning but after 10 minutes it still hasn't finished
(your psq tables don't seem computer generated, did you write them yourself?)

I don't think it's his evaluation, he's clocking 270,000,000 nodes per second. Wondering what method he's using, mailboxes or bitboards?

whoops, 7 million, not too impressive, but much faster than mine.

[hgm]

I'm testing my new engine which is based on mailbox ideas; however, it's not really a mailbox in a formal sense. It's 21x faster than my old inefficient engine, but, I don't have any idea how it compares with a well structured mailbox or bitboard program. So, I tested it on the FEN below with the following statistics:

FEN[201] = "5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1 "; // Unknown author 7-move mate [Qe4]

engine calls: 847,550,776
nodes visited: 34,141233,773
time: 11,327.6 seconds

methods: alpha/beta, history by ply, Zobirst hash, optimized evaluation function: history, capture, check, slider piece, enemy king moves[feedback[], and pawn promotion.

computer chip Intel I5, 2.5 GHz, 16GB ram

Need somebody to compare with.

My engine doesn't have history heuristics yet, nor any prunings except alpha/beta itself. It finds mate in 7 at depth 10:

Code: Select all

go
info score cp 822 depth 1 seldepth 6 time 0 nodes 274 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 822 depth 2 seldepth 7 time 0 nodes 886 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 872 depth 3 seldepth 10 time 3 nodes 4935 nps 1645000 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7c8 f7h7
info score cp 1098 depth 4 seldepth 11 time 7 nodes 23304 nps 3329143 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1098 depth 5 seldepth 13 time 20 nodes 119780 nps 5989000 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7f6 d7f7 f6f7 d3e3
info score cp 1170 depth 6 seldepth 15 time 93 nodes 650253 nps 6991968 hashfull 1 pv c4e4 e3h6 d1d8 f8g7 e4e8 h6f4 e8g8 g7h6 g8f7
info score cp 1264 depth 7 seldepth 17 time 403 nodes 2871969 nps 7126474 hashfull 5 pv d1d8 f8e7 d8c8 f7f8 c4e4 e7f7 c8c7 f7g8 e4e3 f8f7 c7f7 g8f7 e3a7
info score cp 1312 depth 8 seldepth 21 time 1696 nodes 12136697 nps 7156071 hashfull 22 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d6 c7b7 c6c5 c4d3 e3d4 c3d4
info score cp 1549 depth 9 seldepth 21 time 7216 nodes 50730558 nps 7030288 hashfull 95 pv d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 f7b7 a7a5 b7h7
info score mate 7 depth 10 seldepth 23 time 38200 nodes 270719243 nps 7086891 hashfull 399 pv c4e4 e3h6 d1d8 f8g7 f3d4 f7f8 e4e7 g7g8 d8d7 f8f7 e7f7 g8h8 f7h7

38 seconds on a Core i7-6700K.

Single thread?

[tcusr]

i have 3 millions NPS so he should be more than two times faster than me but instead it took me 6 minutes for a search at depth 10

[jonkr]

Code: Select all

SlowChess Blitz 2.81 avx2  (64 MB hash, 1 thread)
FEN: 5k2/ppp2r1p/2p2ppP/8/2Q5/2P1bN2/PP4P1/1K1R4 w - - 0 1

[19]  M7  00:00.5  659K  Qe4  Bxh6  Rd8+  Kg7  Nd4  Bf4  Ne6+  Kh6  Rd1  g5  Rh1+  Bh2  Rxh2#  
[18]  M7  00:00.5  631K  Qe4  Bxh6  Rd8+  Kg7  Nd4  Bf4  Ne6+  Kh6  Rd1  g5  Rh1+  Bh2  Rxh2#  
[18]  M9  00:00.5  603K  Qe4 [++] 
[15]  M10  00:00.2  246K  Rd8+  Ke7  Qd3  f5  Rh8  Bc5  Qe2+  Kd7  Ne5+  Ke6  Nxf7+  Kxf7  Qe5  Bd4  cxd4  c5  Rxh7+  Kg8  Qh8#  
[13]  M12  00:00.0  120K  Rd8+  Ke7  Qd3  f5  Rh8  Bc5  Qd8+  Ke6  Ng5+  Ke5  Nxf7+  Kf4  Qxc7+  Ke3  Qg3+  Kd2  Rd8+  Bd6  Rxd6+  Ke2  Qf3+  Ke1  Rd1#  
[12]  M13  00:00.0  86K  Rd8+  Ke7  Qd3  f5  Rh8  Bc5  Qd8+  Ke6  Ng5+  Ke5  Nxf7+  Ke6  Qxc7  Be7  Qxb7  f4  Re8  a5  Qxe7+  Kd5  Qd6+  Kc4  Qxc6+  Kd3  Qd5#  <tt> 
[12]  M17  00:00.0  80K  Rd8+ [++] 

Slow Chess takes < 0.1 seconds for mate and 0.5 seconds for M7.

This is quite a lopsided position, in multi-pv mode this is what I get :

Code: Select all

1. [26] M7   Qe4  Bxh6  Rd8+  Kg7  Nd4  Bf4  Ne6+  Kh6  Rd1  g5  Rh1+  Bh2  Rxh2#  
2. [26] M8   Qe6  Bxh6  Ne5  fxe5  Rd8+  Kg7  Qxe5+  Rf6  Rd7+  Kg8  Qxf6  c5  Qf7+  Kh8  Qxh7#  
3. [26] M8   Ng5  Bxg5  Qb4+  Ke8  Re1+  Kd8  Qe4  Rf8  Qe7+  Kc8  Qxf8+  Kd7  Qe7+  Kc8  Qe8#  
4. [26] M8   Rd8+  Ke7  Rc8  f5  Rxc7+  Kd6  Qxf7  Kc5  Rxb7  a5  b4+  axb4  cxb4+  Kd6  Qd7#  
5. [26] M9   Nd4  f5  Nxc6  bxc6  Qe6  Bxh6  Rd8+  Kg7  Qe5+  Rf6  Rd7+  Kg8  Qxf6  c5  Qf7+  Kh8  Qxh7#  
6. [26] M10   g4  Bxh6  Rd8+  Kg7  Qe6  Bf4  Qe8  f5  Ne5  Rf6  Qh8+  Kh6  Qxf6  Bxe5  Qh4+  Kg7  Rd7+  Kg8  Qd8#  
7. [26] M10   Qb4+  c5  Qe4  Bxh6  Rd8+  Kg7  Qe8  Bg5  g4  f5  Nxg5  Kf6  Nxf7  fxg4  Qe5+  Kxf7  Rd7+  Kg8  Qe8#  
8. [26] M10   Qg4  Bxh6  Qc8+  Kg7  Rd8  f5  Rg8+  Kf6  Qd8+  Re7  Re8  Bf8  Rxf8+  Kg7  Qxe7+  Kh6  Rh8  Kh5  Qg5#  
9. [26] M10   g3  Bxh6  Nd4  Rd7  Ne6+  Ke7  Rxd7+  Kxd7  Nc5+  Kd6  Qd4+  Ke7  Qd7+  Kf8  Ne6+  Kg8  Qe8+  Bf8  Qxf8#  
10. [26] M11   Qe2  Re7  Rd8+  Re8  Rxe8+  Kxe8  Qxe3+  Kd7  Nd4  c5  Qe6+  Kd8  Qxf6+  Kc8  Qf8+  Kd7  Qf7+  Kd8  Ne6+  Kc8  Qe8# 
 

[mvanthoor]

Single thread?

Yes. Why?

PS: Node counting is done as soon as alpha-beta is entered. Each node is counted, even if it is cut short due to a hash cut. So Rustic counts every node it enters, even if it happens to not generate moves in it. QSearch also counts nodes directly at the beginning. To make sure that leaf nodes at depth == 0 are not counted twice (once in alpha-beta, then again in qsearch), I do a node decrement just before entering qsearch. So it basically undoes the node count for alpha-beta when it happens to be depth 0. (I assume that this is faster than having an extra if-statement to see if depth >= 1.)

In case someone wonders about the 7 million NPS; this is how that count is obtained.

[mvanthoor]

you must have a really good evaluation function then, i only have mvv/lva ordering and delta pruning but after 10 minutes it still hasn't finished
(your psq tables don't seem computer generated, did you write them yourself?)

Thanks. If you are looking at the master branch, then yes. That's still Rustic Alpha 3. I wrote those PST's myself.

The output above is from the development version of Rustic 4, which uses a Texel-tuned tapered evaluation. This version is 300 Elo stronger than version Alpha 3.0.0. The Texel-tuned tapered evaluation is the only chess-playing addition to Rustic between version Rustic Alpha 3 and Rustic 4. The other feature that will be in version 4 is the XBoard protocol, and some code cleanup. 250 out of the 300 Elo comes from the tapered/tuned evaluation.

I could run a test on this position with version Alpha 3, but I assume it will be much worse.

Edit: correction... It is worse, but not by much. Rustic Alpha 3 finds the mate in 7 on depth 10 as well, in 45 seconds:

Code: Select all

info score cp 780 depth 1 seldepth 6 time 0 nodes 227 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 780 depth 2 seldepth 7 time 0 nodes 644 nps 0 pv c4b4 c6c5 b4b7 e3h6 b7a7
info score cp 790 depth 3 seldepth 10 time 2 nodes 3995 nps 1997500 pv d1d8 f8e7 d8d7 e7d7 c4f7 d7c8 f7h7
info score cp 1040 depth 4 seldepth 10 time 6 nodes 21989 nps 3664833 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7e6 d7f7 e6f7 d3e3
info score cp 1040 depth 5 seldepth 13 time 18 nodes 122497 nps 6805389 pv d1d8 f8e7 c4d3 f6f5 d8d7 e7e6 d7f7 e6f7 d3e3
info score cp 1080 depth 6 seldepth 15 time 84 nodes 645139 nps 7680226 hashfull 1 pv d1d8 f8e7 d8c8 e3b6 c4e2 e7d6 c8d8 f7d7 e2d1 b6d4 d8d7 d6d7 d1d4
info score cp 1200 depth 7 seldepth 17 time 403 nodes 3151063 nps 7819015 hashfull 5 pv d1d8 f8e7 d8c8 e3f4 f3d4 e7d6 c4f7 f4h6 c8c7
info score cp 1275 depth 8 seldepth 17 time 1761 nodes 13683641 nps 7770381 hashfull 24 pv d1d8 f8e7 d8c8 f7f8 c8c7 e7d6 c7b7 c6c5 c4d3 e3d4 c3d4
info score cp 1545 depth 9 seldepth 21 time 8151 nodes 61209988 nps 7509507 hashfull 110 pv d1d8 f8e7 d8c8 f6f5 c8c7 e7d6 c7f7 f5f4 f7b7 c6c5 b7a7
info score mate 7 depth 10 seldepth 23 time 45935 nodes 351997491 nps 7662947 hashfull 474 pv c4e4 e3h6 d1d8 f8g7 f3d4 f7f8 e4e7 g7g8 d8d7

I don't know why it searches even more nodes/second; up to 7.8 million even. This version doesn't even count every node; it counts AFTER the hash cuts, so if a node is cut short, it's not counted.

[mvanthoor]

I don't think it's his evaluation, he's clocking 270,000,000 nodes per second. Wondering what method he's using, mailboxes or bitboards?

Just "standard" fancy magic bitboards...

whoops, 7 million, not too impressive, but much faster than mine.

In a few weeks I'll see how much NPS it'll do on an AMD 5700G, which is going to be Rustic's test system. It's about three times as fast. Passmark is ~24.500, where that of the old i7 is about ~8.500. The 5700G has 8 cores though, and the i7 has four, so I assume we can count roughly ~3050 points per core for the 5700G, and 2.125 points for the i7. If the 5700G is indeed 1,44x as fast as the i7-6700K (per core), I expect the engine to reach about 10 million NPS on the 5700G.

However, while speed is important, Rustic still has some way to go with regard to pruning and evaluation. Especially pruning; it can't really get beyond depth 10 middle games at the moment.