Tornado News

[Daniel Shawul]

At the moment i am very pissed off

There are some trolls around that complains everything what programmers are doing, even they are spend many years at own work and let the community enjoy for free !

Hi Engin,
Your anger is very understandable.
I have another implementation that is not GPL but currently it is not compatible with SF nets.
https://github.com/dshawul/nncpu-probe
It uses vertical symmetry to reduce the number of input nodes by half. What I suggest is you simply keep
on training your own net. By that time either I would finish support for SF format nets or you would have implemented
your own nnue prober anyway.

Daniel

i never understand why the input must be so big, i am using just 64 input neurons (for pieces from 1 - 6, black has negative ones) for the whole board and just only one hidden layer with 32 neurons, the output has only 1 neuron that output from -1 to +1 score from white view of point. that is enough and should be very fast calculate on the cpu

if your nn_probe is only probing so how i can training it without a function that can backpropagate the network ??

NNUE's trick is its wide input layer that takes into consideration the position of kings. Think of it like a big piece square table of PSQT[col][square][king-square] instead of the regular PSQT[col][square]. It captures king attack and king defence very well -- which is the main theme of chess evaluation functions especially Stockfish's. My guess is you probably won't get to the level of NNUE with your architecture. With a convolutional neural network, convolutions should be able to capture attacks on kings or otherwise without the need for such a big wide layer. NNUE input layer overfits the data and probably does a lot of memorization for attacks/defending on given king square without really generalizing that knowldege to other squares of the king. What I mean is if it learned a good king attack strategy for a king on E8 it probably doesn't transfer that knowledge to a king on B8 because it doesn't have convolutions or many layers and other infrastrucre needed for generalization of learning. I think Giraffee had to feed attack and defend squares as inputs for this purpose.

The current nncpu code does only a NNUE architecture of Scorpio's format, so for other types of nets you will have to write your own. There is an associated tensorflow training code for that too. I wish tensorflow c++ api had low overheads so that one avoids writing probing code for different neural network architectures. Right now tensorflow is like 300x slower than the hand-written NNUE probing code simply because tensorflow had an overhead of 20ms per call or so. I suspect a convolutional neural network of tiny size should be able to compete with NNUE architecture.

[Engin]

At the moment i am very pissed off

There are some trolls around that complains everything what programmers are doing, even they are spend many years at own work and let the community enjoy for free !

Hi Engin,
Your anger is very understandable.
I have another implementation that is not GPL but currently it is not compatible with SF nets.
https://github.com/dshawul/nncpu-probe
It uses vertical symmetry to reduce the number of input nodes by half. What I suggest is you simply keep
on training your own net. By that time either I would finish support for SF format nets or you would have implemented
your own nnue prober anyway.

Daniel

i never understand why the input must be so big, i am using just 64 input neurons (for pieces from 1 - 6, black has negative ones) for the whole board and just only one hidden layer with 32 neurons, the output has only 1 neuron that output from -1 to +1 score from white view of point. that is enough and should be very fast calculate on the cpu

if your nn_probe is only probing so how i can training it without a function that can backpropagate the network ??

NNUE's trick is its wide input layer that takes into consideration the position of kings. Think of it like a big piece square table of PSQT[col][square][king-square] instead of the regular PSQT[col][square]. It captures king attack and king defence very well -- which is the main theme of chess evaluation functions especially Stockfish's. My guess is you probably won't get to the level of NNUE with your architecture. With a convolutional neural network, convolutions should be able to capture attacks on kings or otherwise without the need for such a big wide layer. NNUE input layer overfits the data and probably does a lot of memorization for attacks/defending on given king square without really generalizing that knowldege to other squares of the king. What I mean is if it learned a good king attack strategy for a king on E8 it probably doesn't transfer that knowledge to a king on B8 because it doesn't have convolutions or many layers and other infrastrucre needed for generalization of learning. I think Giraffee had to feed attack and defend squares as inputs for this purpose.

The current nncpu code does only a NNUE architecture of Scorpio's format, so for other types of nets you will have to write your own. There is an associated tensorflow training code for that too. I wish tensorflow c++ api had low overheads so that one avoids writing probing code for different neural network architectures. Right now tensorflow is like 300x slower than the hand-written NNUE probing code simply because tensorflow had an overhead of 20ms per call or so. I suspect a convolutional neural network of tiny size should be able to compete with NNUE architecture.

nope...my mean square error is now going constantly down, so i get in some hours/days to near zero, a convolutional network is much more to complex and maybe to slow too.

i also tried tensorflow and pytorch for training networks but they are works with python only, i hate to work witch python why the hell all working with python and not with c++ what is much faster even the code of tensorflow is written in cpp code, but there is no easy way to using that in C/C++ ?

[Wilson]

Would it be possible to compile nnue-probe as a dll? Would that solve the GPL issue?

[Engin]

Tornado website is online again !

here: https://sites.google.com/view/tornadochess/start

nothing news since last year december 2020 but i am try to generated my own neural network next or with tuned eval parameters !

[Engin]

My Tornado UCI chess engine is online again !

website: https://sites.google.com/view/tornadochess/start
facebook: https://www.facebook.com/profile.php?id=61553365500953

[chessica]

My Tornado UCI chess engine is online again !

website: https://sites.google.com/view/tornadochess/start
facebook: https://www.facebook.com/profile.php?id=61553365500953
[d]5k2/pp3N2/5K2/5B2/8/8/8/8 w - - 0 1 #9

TornadoChess_8_x64:

Code: Select all

 1/1	00:00,115	 2	0	+6,13	Sf7-e5
 1/1	00:00,116	 22	0	+6,13	Sf7-e5
 2/2	00:00,116	 30	0	+5,46	Sf7-e5 a7-a5
 2/2	00:00,116	 45	0	+6,05	Sf7-d6 a7-a5 Sd6xb7
 2/2	00:00,116	 81	0	+6,05	Sf7-d6 a7-a5 Sd6xb7
 3/3	00:00,117	 131	0	+6,05	Sf7-d6 a7-a5 Sd6xb7
 3/4	00:00,117	 199	0	+6,05	Sf7-d6 a7-a5 Sd6xb7
 4/4	00:00,117	 254	0	+5,60	Sf7-d6 a7-a5 Sd6xb7 a5-a4
 4/6	00:00,117	 396	0	+5,60	Sf7-d6 a7-a5 Sd6xb7 a5-a4
 5/5+	00:00,117	 417	0	+5,68	Sf7-d6 a7-a5 Sd6xb7 a5-a4 Sb7-c5
 5/5+	00:00,118	 432	0	+5,76	Sf7-d6 a7-a5 Sd6xb7 a5-a4 Sb7-c5
 5/5+	00:00,118	 458	0	+5,88	Sf7-d6 a7-a5 Sd6xb7 a5-a4 Sb7-c5
 5/5-	00:00,118	 586	0	+5,52	Sf7-d6 b7-b6 Lf5-e4 a7-a5
 5/5	00:00,118	 685	0	+5,52	Sf7-d6 b7-b6 Sd6-c4 b6-b5 Sc4-e5
 5/8	00:00,119	 1k	0	+5,52	Sf7-d6 b7-b6 Sd6-c4 b6-b5 Sc4-e5
 6/6+	00:00,119	 1k	0	+5,60	Sf7-d6 b7-b6 Sd6-c8 a7-a5 Sc8xb6
 6/6-	00:00,120	 1k	0	+5,44	Sf7-d6 b7-b6 Sd6-c8 b6-b5 Sc8xa7 b5-b4
 6/10	00:00,121	 2k	0	+5,34	Sf7-d6 b7-b6 Sd6-c8 b6-b5 Sc8xa7 b5-b4
 7/8+	00:00,121	 2k	0	+5,42	Sf7-d6 b7-b6 Sd6-c8 b6-b5 Sc8xa7 b5-b4 Sa7-c6
 7/8+	00:00,121	 2k	0	+5,50	Sf7-d6 b7-b6 Sd6-c8 b6-b5 Sc8xa7 b5-b4 Sa7-c6
 7/12	00:00,124	 4k	0	+6,24	Sf7-d6 b7-b6 Sd6-c8 b6-b5 Sc8xa7 b5-b4 Sa7-c6
 8/12-	00:00,124	 4k	0	+6,16	Sf7-d6 b7-b6 Sd6-c8 b6-b5 Sc8xa7 b5-b4 Sa7-c6 b4-b3
 8/12-	00:00,125	 4k	0	+6,08	Sf7-d6 b7-b6 Sd6-c8 b6-b5 Sc8xa7 b5-b4 Sa7-c6 b4-b3
 8/14	00:00,128	 6k	0	+5,85	Sf7-d6 b7-b6 Sd6-c8 b6-b5 Sc8xa7 b5-b4 Sa7-c6 b4-b3
 9/14+	00:00,130	 7k	0	+5,93	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8
 9/14+	00:00,131	 7k	0	+6,01	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8
 9/14+	00:00,131	 7k	0	+6,13	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8
 9/14	00:00,132	 8k	0	+6,18	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 a5-a4 Lb5xa4
 9/15	00:00,138	 11k	0	+6,18	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 a5-a4 Lb5xa4
 10/12+	00:00,139	 12k	0	+6,26	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 a5-a4 Lb5xa4
 10/12+	00:00,139	 12k	0	+6,34	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 a5-a4 Lb5xa4
 10/13+	00:00,140	 13k	0	+6,46	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 a5-a4 Lb5xa4
 10/16	00:00,143	 15k	0	+6,47	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 10/16	00:00,149	 20k	0	+6,47	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 11/15+	00:00,151	 21k	0	+6,55	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 11/18+	00:00,155	 24k	0	+6,63	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 11/18+	00:00,160	 28k	0	+6,93	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 11/18	00:00,171	 35k	0	+7,03	Sf7-d6 b7-b5 Sd6xb5 a7-a5 Lf5-c2 Kf8-e8 Sb5-d6+ Ke8-d7 Sd6-c4 Kd7-c6
 11/18	00:00,181	 44k	0	+7,03	Sf7-d6 b7-b5 Sd6xb5 a7-a5 Lf5-c2 Kf8-e8 Sb5-d6+ Ke8-d7 Sd6-c4 Kd7-c6
 12/18+	00:00,190	 49k	0	+7,11	Sf7-d6 b7-b5 Sd6xb5 a7-a5 Lf5-d7 Kf8-g8 Sb5-d6 Kg8-f8 Sd6-c4
 12/18+	00:00,191	 51k	0	+7,19	Sf7-d6 b7-b5 Sd6xb5 a7-a5 Lf5-d7 Kf8-g8 Sb5-d6 Kg8-f8 Sd6-c4
 12/19+	00:00,197	 55k	0	+7,49	Sf7-d6 b7-b5 Sd6xb5 a7-a5 Lf5-d7 Kf8-g8 Sb5-d6 Kg8-f8 Sd6-c4
 12/19	00:00,204	 60k	0	+7,12	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 12/21	00:00,222	 73k	0	+7,12	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 13/18+	00:00,224	 75k	0	+7,20	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 13/18+	00:00,227	 77k	0	+7,28	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 13/20+	00:00,229	 80k	0	+7,40	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 13/21+	00:00,239	 87k	0	+7,58	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-f8
 13/21+	00:00,250	 96k	0	+7,85	Sf7-d6 a7-a6 Sd6xb7 Kf8-e8 Lf5-e6 Ke8-f8 Sb7-d6 a6-a5 Le6-d7 Kf8-g8
 13/24	00:00,263	 106k	0	+7,70	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-f8
 13/24	00:00,410	 128k	0	+7,70	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-f8
 14/22+	00:00,421	 136k	0	+7,78	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-f8
 14/22+	00:00,422	 138k	0	+7,86	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-f8
 14/22+	00:00,442	 154k	0	+7,98	Sf7-d6 a7-a5 Lf5-d7 b7-b6 Sd6-c4 Kf8-g8 Sc4xb6 Kg8-h7 Sb6-c4 a5-a4
 14/27	00:00,469	 174k	0	+7,72	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-h7
 14/27	00:00,515	 209k	0	+7,72	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-h7
 15/25	00:00,622	 293k	0	+7,71	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 Kf8-g8 Lf5-d3 Kg8-h8 Sd6-c4 Kh8-g8
 15/25	00:00,709	 359k	0	+7,71	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 Kf8-g8 Lf5-d3 Kg8-h8 Sd6-c4 Kh8-g8
 16/22+	00:00,719	 368k	0	+7,79	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 Kf8-g8 Lf5-d3 Kg8-h8 Sd6-f7+ Kh8-g8
 16/23+	00:00,726	 374k	0	+7,87	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 Kf8-g8 Lf5-d3 Kg8-h8 Sd6-f7+ Kh8-g8
 16/23+	00:00,739	 384k	0	+7,99	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-h8 Sb7-d6 Kh8-g8 Lf5-e4 a7-a5 Sd6-c4 a5-a4
 16/25	00:00,810	 440k	0	+7,68	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 16/26	00:00,899	 521k	0	+7,68	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 17/24+	00:00,917	 537k	0	+7,76	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 17/24+	00:00,925	 544k	0	+7,84	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 17/25+	00:00,934	 551k	0	+7,96	Sf7-d6 b7-b6 Lf5-d3 a7-a5 Ld3-b5 Kf8-g8 Sd6-c8 Kg8-h7 Sc8xb6 Kh7-h6
 17/26+	00:00,966	 579k	0	+8,14	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 a7-a6 Lb5xa6 Kf8-e8
 17/32	00:01,106	 701k	701k	+8,28	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-h7
 17/32	00:01,254	 832k	832k	+8,28	Sf7-d6 b7-b6 Lf5-d3 Kf8-g8 Sd6-c8 Kg8-f8 Ld3-b5 Kf8-g8 Sc8xa7 Kg8-h7
 18/27+	00:01,435	 999k	999k	+8,36	Sf7-d6 b7-b6 Lf5-e6 a7-a6 Le6-f7 b6-b5 Sd6-e4 a6-a5 Se4-c5
 18/32+	00:01,595	 1.149k	1.149k	+8,44	Sf7-d6 b7-b6 Lf5-e6 a7-a6 Le6-f7 b6-b5 Sd6-e4 a6-a5 Se4-c5
 18/32+	00:01,679	 1.229k	1.229k	+8,56	Sf7-d6 b7-b6 Lf5-e6 a7-a6 Le6-f7 b6-b5 Sd6-e4 a6-a5 Se4-c5
 18/32	00:01,829	 1.365k	1.365k	+8,55	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 a7-a5 Lf5-g6 a5-a4 Sd6-c4 Kf8-g8
 18/32	00:01,962	 1.480k	1.480k	+8,55	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 a7-a5 Lf5-g6 a5-a4 Sd6-c4 Kf8-g8
 19/30+	00:02,085	 1.595k	797k	+8,63	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 a7-a5 Lf5-g6 a5-a4 Sd6-c4 Kf8-g8
 19/30+	00:02,107	 1.616k	808k	+8,71	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 Kf8-g8 Lf5-e4 a7-a5 Le4-c2 Kg8-f8
 19/35	00:02,680	 2.139k	1.070k	+8,75	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 Kf8-g8 Lf5-e4 a7-a5 Le4-c2 Kg8-f8
 20/30+	00:02,716	 2.176k	1.088k	+8,83	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 Kf8-g8 Lf5-e4 a7-a5 Le4-c2 a5-a4
 20/30+	00:02,724	 2.182k	1.091k	+8,91	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 Kf8-g8 Lf5-e4 a7-a5 Le4-d5+ Kg8-h8
 20/35	00:03,185	 2.626k	875k	+8,91	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 a7-a5 Lf5-c2 Kf8-g8 Lc2-b3+ Kg8-h7
 20/46	00:03,780	 3.155k	1.052k	+8,91	Sf7-d6 Kf8-g8 Sd6xb7 Kg8-f8 Sb7-d6 a7-a5 Lf5-c2 Kf8-g8 Lc2-b3+ Kg8-h7
 21/36-	00:04,554	 3.905k	976k	+8,83	Sf7-d6 a7-a5 Lf5-e6 a5-a4 Le6-f7 a4-a3 Sd6xb7 a3-a2 Lf7xa2 Kf8-e8
 21/36-	00:04,859	 4.195k	1.049k	+8,75	Sf7-d6 a7-a5 Lf5-e6 a5-a4 Le6-f7 a4-a3 Sd6xb7 a3-a2 Lf7xa2 Kf8-e8
 21/36	00:04,997	 4.326k	1.082k	+8,92	Sf7-d6 a7-a5 Lf5-e6 a5-a4 Le6-f7 a4-a3 Sd6xb7 a3-a2 Lf7xa2 Kf8-e8
 21/36	00:06,030	 5.301k	883k	+8,98	Lf5-d7 b7-b6 Ld7-b5 a7-a6 Lb5xa6 b6-b5 La6xb5 Kf8-g8 Sf7-d6 Kg8-f8
 21/36	00:06,045	 5.313k	885k	+8,98	Lf5-d7 b7-b6 Ld7-b5 a7-a6 Lb5xa6 b6-b5 La6xb5 Kf8-g8 Sf7-d6 Kg8-f8
 22/38	00:06,944	 6.225k	1.037k	+9,00	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6xb7 a5-a4 Ld7xa4 Kf8-g8 Sb7-d8 Kg8-h7
 22/38	00:08,294	 7.519k	940k	+9,00	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6xb7 a5-a4 Ld7xa4 Kf8-g8 Sb7-d8 Kg8-h7
 23/38	00:09,388	 8.677k	964k	+9,00	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6xb7 a5-a4 Ld7xa4 Kf8-g8 Sb7-d8 Kg8-h7
 23/39	00:10,823	 10.095k	1.010k	+9,00	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6xb7 a5-a4 Ld7xa4 Kf8-g8 Sb7-d8 Kg8-h7
 24/37	00:12,243	 11.634k	969k	+9,00	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6xb7 a5-a4 Ld7xa4 Kf8-g8 Sb7-d8 Kg8-h7
 24/38	00:14,836	 14.172k	1.012k	+9,00	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6xb7 a5-a4 Ld7xa4 Kf8-g8 Sb7-d8 Kg8-h7
 25/40	00:16,324	 15.822k	989k	+9,06	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6xb7 a5-a4 Ld7xa4 Kf8-g8 Sb7-d8 Kg8-h7
 25/45	00:20,222	 19.659k	983k	+9,06	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6xb7 a5-a4 Ld7xa4 Kf8-g8 Sb7-d8 Kg8-h7
 26/45+	00:24,189	 23.934k	997k	+9,14	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6-b5 b7-b6 Sb5-d4 Kf8-g8 Ld7-e6+ Kg8-h8
 26/45+	00:24,189	 23.934k	997k	+9,22	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6-b5 b7-b6 Sb5-d4 Kf8-g8 Ld7-e6+ Kg8-h8
 26/45	00:24,206	 23.942k	998k	+112,36	Lf5-d7 a7-a6 Sf7-d6 a6-a5 Sd6-b5 b7-b6 Sb5-d4 Kf8-g8 Ld7-e6+ Kg8-h8
 27/45	00:29,713	 29.378k	1.013k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 27/45	00:32,246	 32.038k	1.001k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 28/28	00:32,656	 32.520k	1.016k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 28/28	00:34,152	 34.144k	1.004k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 29/29	00:34,742	 34.852k	1.025k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 29/29	00:37,292	 37.828k	1.022k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 30/30	00:37,954	 38.684k	1.046k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 30/30	00:41,021	 42.366k	1.033k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 31/31	00:41,931	 43.543k	1.062k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 31/31	00:45,446	 47.785k	1.062k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8
 32/32	00:46,429	 49.090k	1.067k	+M8	Lf5-d7 a7-a6 Sf7-g5 b7-b5 Sg5-h7+ Kf8-g8 Kf6-g6 b5-b4 Ld7-e6+ Kg8-h8

[chessica]

Old PC with sse2 popcnt

Here, I get no engine output


2023-11-09 20:08:50,564-->1:stop
2023-11-09 20:08:57,840-->1:quit
2023-11-09 20:08:58,256*1*-----------------------Starte Motor 1 Tornado-9-NNUE------------------------
2023-11-09 20:08:58,256*1*Configured Engine 1 Type: UCI
2023-11-09 20:08:58,256*1*Engine 1 dir: C:\Users\Privat\Desktop\Tornado NNUE
2023-11-09 20:08:58,256*1*Engine 1 commandline: "C:\Users\Privat\Desktop\Tornado NNUE\Tornado.exe"
2023-11-09 20:08:58,371*1*Child Process Prio Adj: PID 2052 conhost.exe
2023-11-09 20:08:58,426*1*Engine 1 ProcessID: 1868
2023-11-09 20:08:58,426*1*Engine 1 Prio:32 ThreadPrio:0
2023-11-09 20:08:58,456-->1:uci
2023-11-09 20:08:58,596<--1:Loading NNUE : nnue.bin
2023-11-09 20:08:58,641<--1:NNUE loaded !
2023-11-09 20:08:58,641<--1:Tornado Dec 9 2020 x64
2023-11-09 20:08:58,641<--1:by Engin Uestuen
2023-11-09 20:08:58,641<--1:id name Tornado Dec 9 2020 x64
2023-11-09 20:08:58,641<--1:id author Engin Uestuen
2023-11-09 20:08:58,641<--1:option name UCI_EngineAbout type string default Tornado Dec 9 2020 x64, by Engin Uestuen
2023-11-09 20:08:58,641<--1:option name Hash type spin default 256 min 1 max 32768
2023-11-09 20:08:58,641<--1:option name EvalHashPerThread type spin default 128 min 1 max 256
2023-11-09 20:08:58,641<--1:option name PawnHashPerThread type spin default 64 min 1 max 256
2023-11-09 20:08:58,641<--1:option name MaterialHashPerThread type spin default 32 min 1 max 256
2023-11-09 20:08:58,641<--1:option name ClearHash type button
2023-11-09 20:08:58,641<--1:option name Ponder type check default true
2023-11-09 20:08:58,656<--1:option name UCI_Chess960 type check default false
2023-11-09 20:08:58,656<--1:option name Castling_OO type check default false
2023-11-09 20:08:58,656<--1:option name Threads type spin default 1 min 1 max 32
2023-11-09 20:08:58,656<--1:option name NalimovPath type string default C:\nalimov\
2023-11-09 20:08:58,656<--1:option name NalimovCache type spin default 1 min 1 max 32
2023-11-09 20:08:58,656<--1:option name UseNalimov type check default true
2023-11-09 20:08:58,656<--1:option name DebugInfoString type check default false
2023-11-09 20:08:58,656<--1:option name SSE_Popcount type check default true
2023-11-09 20:08:58,656<--1:option name NNUE_Eval type check default true
2023-11-09 20:08:58,656<--1:option name NNUE_File type string default nnue.bin
2023-11-09 20:08:58,656<--1:uciok
2023-11-09 20:08:58,656-->1:setoption name Hash value 128
2023-11-09 20:08:58,656-->1:setoption name NalimovPath value C:\Program Files\arena_3.5.1\TB\Nalimov\TBS1
2023-11-09 20:08:58,656-->1:setoption name NalimovCache value 32
2023-11-09 20:08:58,671*1*Child Process Prio Adj: PID 2052 conhost.exe
2023-11-09 20:08:58,686-->1:isready
2023-11-09 20:08:58,811<--1:readyok
2023-11-09 20:09:15,389**Dauer Start Motoren für Analyse:0 ms
2023-11-09 20:09:15,389-->1:isready
2023-11-09 20:09:15,389<--1:readyok
2023-11-09 20:09:15,404-->1:ucinewgame
2023-11-09 20:09:15,404-->1:isready
2023-11-09 20:09:15,504-->1:position fen 5k2/pp3N2/5K2/5B2/8/8/8/8 w - - 0 1
2023-11-09 20:09:15,504-->1:go infinite
2023-11-09 20:09:15,504**Dauer Start Analyse:125
2023-11-09 20:09:15,558<--1:readyok
2023-11-09 20:09:36,442-->1:stop

[Engin]

Old PC with sse2 popcnt

Here, I get no engine output


2023-11-09 20:08:50,564-->1:stop
2023-11-09 20:08:57,840-->1:quit
2023-11-09 20:08:58,256*1*-----------------------Starte Motor 1 Tornado-9-NNUE------------------------
2023-11-09 20:08:58,256*1*Configured Engine 1 Type: UCI
2023-11-09 20:08:58,256*1*Engine 1 dir: C:\Users\Privat\Desktop\Tornado NNUE
2023-11-09 20:08:58,256*1*Engine 1 commandline: "C:\Users\Privat\Desktop\Tornado NNUE\Tornado.exe"
2023-11-09 20:08:58,371*1*Child Process Prio Adj: PID 2052 conhost.exe
2023-11-09 20:08:58,426*1*Engine 1 ProcessID: 1868
2023-11-09 20:08:58,426*1*Engine 1 Prio:32 ThreadPrio:0
2023-11-09 20:08:58,456-->1:uci
2023-11-09 20:08:58,596<--1:Loading NNUE : nnue.bin
2023-11-09 20:08:58,641<--1:NNUE loaded !
2023-11-09 20:08:58,641<--1:Tornado Dec 9 2020 x64
2023-11-09 20:08:58,641<--1:by Engin Uestuen
2023-11-09 20:08:58,641<--1:id name Tornado Dec 9 2020 x64
2023-11-09 20:08:58,641<--1:id author Engin Uestuen
2023-11-09 20:08:58,641<--1:option name UCI_EngineAbout type string default Tornado Dec 9 2020 x64, by Engin Uestuen
2023-11-09 20:08:58,641<--1:option name Hash type spin default 256 min 1 max 32768
2023-11-09 20:08:58,641<--1:option name EvalHashPerThread type spin default 128 min 1 max 256
2023-11-09 20:08:58,641<--1:option name PawnHashPerThread type spin default 64 min 1 max 256
2023-11-09 20:08:58,641<--1:option name MaterialHashPerThread type spin default 32 min 1 max 256
2023-11-09 20:08:58,641<--1:option name ClearHash type button
2023-11-09 20:08:58,641<--1:option name Ponder type check default true
2023-11-09 20:08:58,656<--1:option name UCI_Chess960 type check default false
2023-11-09 20:08:58,656<--1:option name Castling_OO type check default false
2023-11-09 20:08:58,656<--1:option name Threads type spin default 1 min 1 max 32
2023-11-09 20:08:58,656<--1:option name NalimovPath type string default C:\nalimov\
2023-11-09 20:08:58,656<--1:option name NalimovCache type spin default 1 min 1 max 32
2023-11-09 20:08:58,656<--1:option name UseNalimov type check default true
2023-11-09 20:08:58,656<--1:option name DebugInfoString type check default false
2023-11-09 20:08:58,656<--1:option name SSE_Popcount type check default true
2023-11-09 20:08:58,656<--1:option name NNUE_Eval type check default true
2023-11-09 20:08:58,656<--1:option name NNUE_File type string default nnue.bin
2023-11-09 20:08:58,656<--1:uciok
2023-11-09 20:08:58,656-->1:setoption name Hash value 128
2023-11-09 20:08:58,656-->1:setoption name NalimovPath value C:\Program Files\arena_3.5.1\TB\Nalimov\TBS1
2023-11-09 20:08:58,656-->1:setoption name NalimovCache value 32
2023-11-09 20:08:58,671*1*Child Process Prio Adj: PID 2052 conhost.exe
2023-11-09 20:08:58,686-->1:isready
2023-11-09 20:08:58,811<--1:readyok
2023-11-09 20:09:15,389**Dauer Start Motoren für Analyse:0 ms
2023-11-09 20:09:15,389-->1:isready
2023-11-09 20:09:15,389<--1:readyok
2023-11-09 20:09:15,404-->1:ucinewgame
2023-11-09 20:09:15,404-->1:isready
2023-11-09 20:09:15,504-->1:position fen 5k2/pp3N2/5K2/5B2/8/8/8/8 w - - 0 1
2023-11-09 20:09:15,504-->1:go infinite
2023-11-09 20:09:15,504**Dauer Start Analyse:125
2023-11-09 20:09:15,558<--1:readyok
2023-11-09 20:09:36,442-->1:stop

i think you need a least like a PC with SSE 4 support !

[chessica]

W/o sources I can`t use it.

[Engin]

W/o sources I can`t use it.

sorry for that, i didnt wrote the NNUE probe, its works on my older PC AMD FX 8350 very fine, just not with AVX2 support