CEGT - rating lists June 30th 2019

[Daniel Shawul]

Hi Werner,

Thanks for the test! This Stoofvless seems too strong for scorpio, or scorpio doesn't perform as well on the GTX 1650.
I have always tested on a Volta so i had no idea of its strength other than on that.
Btw how much nps is leela getting on the GTX 1650? The 2 knps that scorpio is getting seems too low.
I may want to try INT8 which may double nps -- it seems from my tests with equal nps it is only -38 elo weaker
than half precision (FP16). So with nps doubling it may perform better than half precision.

Edit: One more thing I forgot in my setup. Yo might want to try delay 1 and compare its speedup with delay 0.
This is most probably needed on a system with few CPU cores (1 or 2 cores) and when launching 128 threads like we do here.
Please try this first before the INT8 i think it may double nps

This is probably a very important factor. I get about 100x more speedup difference with delay 1 on my system if I restrict
it to using 1 core, unbelivable i know but it is how scorpio is designed lanunching many threads unlike lc0.
I always tested using all 32 cores, so i never had to use delay=1. I use linux taskset to restrict to 1 core.
I don't expect the nps to increase by 100x on your systems but maybe it will double though ...

Here is a run with delay=0, nps is 161 nodes/s

Code: Select all

[07:13 dabdi@hsw221 bin] > taskset -c 0 ./scorpio mt 128 delay 0 go quit
feature done=0
Number of cores 1 of 32
ht 4194304 X 16 = 64.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 1342169600 X 10 = 12799.9 MB
processors [64]
processors [128]
EgbbProbe 4.3 by Daniel Shawul
egbb_cache 4084 X 8216 = 32.0 MB
0 egbbs loaded !      
Loading neural network : ../nets-maddex/net-maddex.uff
nn_cache 131072 X 1552 = 194.0 MB
Loading graph on /gpu:0
0. main_input 7168 = 112 8 8
1. policy_head 1858 = 1858 1 1
2. value_head 1 = 1 1 1
Neural network loaded !	
loading_time = 8s
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
63 3 366 182  d2-d4 Ng8-f6 c2-c4 e7-e6
64 10 521 559  d2-d4 Ng8-f6 c2-c4 e7-e6 Ng1-f3 d7-d5 Nb1-c3
65 34 690 912  d2-d4 Ng8-f6 c2-c4 e7-e6 Ng1-f3 d7-d5 Nb1-c3 Bf8-e7
66 35 861 1161  e2-e4 c7-c5 Ng1-f3 d7-d6 d2-d4 c5xd4 Nf3xd4 Ng8-f6 Nb1-c3 a7-a6
67 35 1116 1693  e2-e4 c7-c5 Ng1-f3 d7-d6 d2-d4 c5xd4 Nf3xd4 Ng8-f6 Nb1-c3 a7-a6 Bc1-e3
68 35 1286 2079  e2-e4 c7-c5 Ng1-f3 d7-d6 d2-d4 c5xd4 Nf3xd4 Ng8-f6 Nb1-c3 a7-a6 Bc1-e3
69 34 1457 2498  e2-e4 c7-c5 Ng1-f3 d7-d6 d2-d4 c5xd4 Nf3xd4 Ng8-f6 Nb1-c3 a7-a6 Bc1-e3 e7-e5 Nd4-b3
70 34 1626 2626  e2-e4 c7-c5 Ng1-f3 d7-d6 d2-d4 c5xd4 Nf3xd4 Ng8-f6 Nb1-c3 a7-a6 Bc1-e3 e7-e5 Nd4-b3

# Move   Value=(V,P,V+P)   Policy  Visits                  PV
#----------------------------------------------------------------------------------
#  1   (0.547,0.527,0.537)  22.82     804   d2-d4 d7-d5 c2-c4 e7-e6 Nb1-c3 Ng8-f6 Ng1-f3 Bf8-e7 g2-g3 Ke8-g8
#  2   (0.549,0.527,0.538)  22.76     657   e2-e4 c7-c5 Ng1-f3 d7-d6 d2-d4 c5xd4 Nf3xd4 Ng8-f6 Nb1-c3 a7-a6 Bc1-e3 e7-e5 Nd4-b3
#  3   (0.539,0.527,0.533)   9.56     292   c2-c4 e7-e5 Nb1-c3 Ng8-f6 Ng1-f3 Nb8-c6 g2-g3 Bf8-b4 Bf1-g2
#  4   (0.533,0.527,0.530)   9.16     279   Ng1-f3 d7-d5 d2-d4 Ng8-f6 c2-c4 e7-e6 Nb1-c3 Bf8-e7 g2-g3 Ke8-g8
#  5   (0.520,0.527,0.520)   4.41     114   g2-g3 d7-d5 Ng1-f3 c7-c5 Bf1-g2 Ng8-f6 Ke1-g1 Nb8-c6 d2-d4
#  6   (0.524,0.527,0.524)   4.03     107   e2-e3 Ng8-f6 d2-d4 d7-d5 c2-c4 e7-e6 Ng1-f3
#  7   (0.497,0.527,0.497)   3.03      45   b2-b3 e7-e5 Bc1-b2 Nb8-c6 e2-e3 Ng8-f6 Ng1-f3 e5-e4
#  8   (0.510,0.527,0.510)   2.83      59   c2-c3 d7-d5 d2-d4 Ng8-f6 Ng1-f3 e7-e6 Bc1-f4
#  9   (0.507,0.527,0.507)   2.69      50   Nb1-c3 d7-d5 d2-d4 Ng8-f6 Bc1-f4 a7-a6 e2-e3 e7-e6 Ng1-f3
# 10   (0.496,0.527,0.496)   2.68      38   d2-d3 d7-d5 Ng1-f3 Ng8-f6 g2-g3 c7-c5 Bf1-g2 Nb8-c6
# 11   (0.458,0.527,0.458)   2.46      24   f2-f4 d7-d5 Ng1-f3 Ng8-f6 e2-e3 c7-c5
# 12   (0.473,0.527,0.473)   2.07      25   b2-b4 e7-e5 Bc1-b2 Bf8xb4 Bb2xe5 Ng8-f6 c2-c3 Bb4-e7 e2-e3
# 13   (0.492,0.527,0.492)   1.93      32   h2-h3 d7-d5 d2-d4 c7-c5 e2-e3 Ng8-f6 Ng1-f3 Nb8-c6
# 14   (0.498,0.527,0.498)   1.89      36   a2-a3 d7-d5 d2-d4 Ng8-f6 Ng1-f3 g7-g6 c2-c4 Bf8-g7
# 15   (0.434,0.527,0.434)   1.60      12   Ng1-h3 d7-d5 g2-g3 e7-e5 d2-d4 e5xd4
# 16   (0.473,0.527,0.473)   1.40      17   a2-a4 e7-e5 e2-e4 Ng8-f6 Nb1-c3 Bf8-b4 Ng1-f3
# 17   (0.453,0.527,0.453)   1.26      12   h2-h4 d7-d5 d2-d4 c7-c5 d4xc5 Ng8-f6
# 18   (0.388,0.527,0.388)   1.24       6   g2-g4 d7-d5 h2-h3 e7-e5 Bf1-g2
# 19   (0.418,0.527,0.418)   1.20       8   f2-f3 e7-e5 Nb1-c3 Ng8-f6 d2-d4 e5xd4
# 20   (0.437,0.527,0.437)   0.99       8   Nb1-a3 e7-e5 e2-e3 d7-d5 d2-d4 e5-e4

# nodes = 23219 <0% qnodes> time = 16270ms nps = 1427 eps = 0 nneps = 178
# Tree: nodes = 3580 depth = 12 pps = 161 visits = 2626 
#       qsearch_calls = 0 search_calls = 0
move e2e4
Bye Bye

And with delay=1 gives 16000 nodes/s

Code: Select all

[07:14 dabdi@hsw221 bin] > taskset -c 0 ./scorpio mt 128 delay 1 go quit
feature done=0
Number of cores 1 of 32
ht 4194304 X 16 = 64.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 1342169600 X 10 = 12799.9 MB
processors [64]
processors [128]
EgbbProbe 4.3 by Daniel Shawul
egbb_cache 4084 X 8216 = 32.0 MB
0 egbbs loaded !      
Loading neural network : ../nets-maddex/net-maddex.uff
nn_cache 131072 X 1552 = 194.0 MB
Loading graph on /gpu:0
0. main_input 7168 = 112 8 8
1. policy_head 1858 = 1858 1 1
2. value_head 1 = 1 1 1
Neural network loaded !	
loading_time = 8s
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
63 36 111 17518  e2-e4 Nb8-c6 d2-d4 d7-d5 e4-e5 Bc8-f5 c2-c3 e7-e6 Nb1-d2 f7-f6 f2-f4 f6xe5 f4xe5
64 36 223 35094  e2-e4 e7-e5 Ng1-f3 Nb8-c6 d2-d4 e5xd4 Nf3xd4 Ng8-f6 Nd4xc6 b7xc6 Bf1-d3 d7-d5 e4xd5 c6xd5 Ke1-g1 Bf8-e7 c2-c4
65 29 334 51851  d2-d4 Ng8-f6 c2-c4 c7-c6 Nb1-c3 d7-d5 Ng1-f3 e7-e6 Bc1-g5 h7-h6 Bg5xf6 Qd8xf6
66 27 446 71390  d2-d4 Ng8-f6 c2-c4 e7-e6 e2-e3 Bf8-e7 Ng1-f3 Ke8-g8 Nb1-c3 d7-d5 b2-b3 c7-c5 d4xc5
67 36 557 90129  e2-e4 e7-e6 d2-d4 d7-d5 Nb1-c3 Ng8-f6 Bc1-g5 d5xe4 Nc3xe4 Bf8-e7 Bg5xf6 Be7xf6 Ng1-f3 Ke8-g8 Qd1-d2 Nb8-d7 Ke1-c1 b7-b6
68 25 669 106481  e2-e4 e7-e6 d2-d4 d7-d5 Nb1-c3 Ng8-f6 Bf1-b5 c7-c6 Bb5-d3 c6-c5 d4xc5 Bf8xc5 Ng1-f3 d5xe4 Nc3xe4 Nf6xe4 Bd3xe4
69 24 781 125734  e2-e4 e7-e6 d2-d4 d7-d5 Nb1-c3 Ng8-f6 Bf1-b5 c7-c6 Bb5-d3 c6-c5 d4xc5 Bf8xc5 Ng1-f3 d5xe4 Nc3xe4 Nf6xe4 Bd3xe4 Qd8xd1 Ke1xd1
70 28 893 143024  d2-d4 Ng8-f6 Ng1-f3 e7-e6 g2-g3 c7-c5 Bf1-g2 c5xd4 Nf3xd4 d7-d5 Ke1-g1 e6-e5 Nd4-b3 Bc8-e6 Nb1-c3
71 28 1004 165996  d2-d4 d7-d5 c2-c4 e7-e6 g2-g3 d5xc4 Bf1-g2 c7-c5 Ng1-f3 Nb8-c6 Qd1-a4 c5xd4 Nf3xd4
72 29 1033 170871  d2-d4 d7-d5 c2-c4 e7-e6 Nb1-c3 a7-a6 c4xd5 e6xd5 Ng1-f3 Ng8-f6 Bc1-g5 Bc8-e6 e2-e3 Nb8-d7 Bf1-d3 Bf8-d6 Ke1-g1

# Move   Value=(V,P,V+P)   Policy  Visits                  PV
#----------------------------------------------------------------------------------
#  1   (0.543,0.527,0.535)  22.82   73528   d2-d4 d7-d5 c2-c4 e7-e6 Nb1-c3 a7-a6 c4xd5 e6xd5 Ng1-f3 Ng8-f6 Bc1-g5 Bc8-e6 e2-e3 Nb8-d7 Bf1-d3 Bf8-d6 Ke1-g1
#  2   (0.537,0.527,0.532)  22.76   60350   e2-e4 e7-e6 d2-d4 d7-d5 Nb1-c3 Ng8-f6 e4-e5 Nf6-d7 f2-f4 a7-a6 Ng1-f3 c7-c5 Bc1-e3 Nb8-c6 Qd1-d2 Bf8-e7 d4xc5 Nd7xc5 Ke1-c1
#  3   (0.520,0.527,0.520)   9.56   13451   c2-c4 e7-e5 g2-g3 Ng8-f6 Bf1-g2 Bf8-c5 Nb1-c3 Ke8-g8 Ng1-f3 Nb8-c6 Ke1-g1 d7-d6 e2-e3 Rf8-e8
#  4   (0.530,0.527,0.529)   9.16   14483   Ng1-f3 d7-d5 d2-d4 Ng8-f6 e2-e3 c7-c5 c2-c4 c5xd4 e3xd4 g7-g6 Nb1-c3 Bf8-g7
#  5   (0.502,0.527,0.502)   4.41    1386   g2-g3 d7-d5 Ng1-f3 c7-c5 Bf1-g2 Ng8-f6 Ke1-g1 Nb8-c6 d2-d4 c5xd4 Nf3xd4 e7-e5
#  6   (0.512,0.527,0.512)   4.03    1680   e2-e3 e7-e6 c2-c4 d7-d5 Ng1-f3 Ng8-f6 d2-d4 Bf8-e7 Bf1-e2
#  7   (0.475,0.527,0.475)   3.03     802   b2-b3 e7-e5 Bc1-b2 Nb8-c6 e2-e3 Ng8-f6 Ng1-f3 e5-e4 Nf3-d4 Bf8-c5 Nd4xc6 d7xc6 d2-d4
#  8   (0.488,0.527,0.488)   2.83     804   c2-c3 d7-d5 d2-d4 Ng8-f6 Ng1-f3 e7-e6 Bc1-f4 Bf8-d6 e2-e3 Bd6xf4
#  9   (0.495,0.527,0.495)   2.69     803   Nb1-c3 d7-d5 d2-d4 Ng8-f6 Bc1-f4 a7-a6 e2-e3 e7-e6 Ng1-f3 c7-c5 Bf1-e2 Nb8-c6 Ke1-g1
# 10   (0.497,0.527,0.497)   2.68     753   d2-d3 d7-d5 Ng1-f3 Ng8-f6 g2-g3 c7-c5 Bf1-g2 Nb8-c6 Ke1-g1 e7-e5 e2-e4
# 11   (0.468,0.527,0.468)   2.46     413   f2-f4 d7-d5 Ng1-f3 Ng8-f6 e2-e3 c7-c5 b2-b3 g7-g6 Bf1-b5
# 12   (0.475,0.527,0.475)   2.07     383   b2-b4 e7-e5 Bc1-b2 Bf8xb4 Bb2xe5 Ng8-f6 c2-c3 Bb4-e7 e2-e3 Ke8-g8 d2-d4
# 13   (0.495,0.527,0.495)   1.93     516   h2-h3 d7-d5 d2-d4 c7-c5 e2-e3 Ng8-f6 Ng1-f3 Nb8-c6 c2-c4 e7-e6
# 14   (0.504,0.527,0.504)   1.89     627   a2-a3 d7-d5 d2-d4 Ng8-f6 Ng1-f3 g7-g6 c2-c4 Bf8-g7 c4xd5
# 15   (0.431,0.527,0.431)   1.60     177   Ng1-h3 d7-d5 g2-g3 e7-e5 d2-d4 e5xd4 Qd1xd4 Nb8-c6
# 16   (0.475,0.527,0.475)   1.40     258   a2-a4 e7-e5 e2-e4 Ng8-f6 Nb1-c3 Bf8-b4 Ng1-f3 Ke8-g8 Nf3xe5
# 17   (0.449,0.527,0.449)   1.26     168   h2-h4 d7-d5 d2-d4 c7-c5 d4xc5 Ng8-f6 Ng1-f3 e7-e6 Bc1-e3
# 18   (0.375,0.527,0.375)   1.24      91   g2-g4 d7-d5 h2-h3 e7-e5 Bf1-g2 Nb8-c6 Nb1-c3 Ng8-e7
# 19   (0.418,0.527,0.418)   1.20     118   f2-f3 e7-e5 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Qd1xd4 Nb8-c6
# 20   (0.444,0.527,0.444)   0.99     124   Nb1-a3 e7-e5 e2-e3 d7-d5 d2-d4 e5-e4 c2-c4 c7-c6

# nodes = 1608289 <0% qnodes> time = 10347ms nps = 155435 eps = 0 nneps = 16556
# Tree: nodes = 236447 depth = 21 pps = 16532 visits = 170916 
#       qsearch_calls = 0 search_calls = 0
move d2d4
Bye Bye

regards,
Daniel

[Werner]

Hi Daniel,
here are the last 3 lines from Arena F4 analyzing 20 sec using 2CPU

Arena analyze 20 sec delay 1
2019-07-01 16:21:00,611<--1:# nodes = 242945 <0% qnodes> time = 4843ms nps = 50164 eps = 0 nneps = 5934
2019-07-01 16:21:00,611<--1:# Tree: nodes = 39796 depth = 19 pps = 5923 visits = 28688
2019-07-01 16:21:00,611<--1:# qsearch_calls = 0 search_calls = 0

Arena analyze 20 sec delay 0
2019-07-01 16:23:10,645<--1:# nodes = 156697 <0% qnodes> time = 1718ms nps = 91208 eps = 0 nneps = 11233
2019-07-01 16:23:10,645<--1:# Tree: nodes = 26852 depth = 18 pps = 11212 visits = 19263
2019-07-01 16:23:10,645<--1:# qsearch_calls = 0 search_calls = 0

or do you want to see the Output in a normal game?

here is the Output delay 0 with all 4 CPUs inside command prompt
# nodes = 172671 <0% qnodes> time = 11013ms nps = 15678 eps = 0 nneps = 1978
# Tree: nodes = 29412 depth = 19 pps = 1957 visits = 21341
# qsearch_calls = 0 search_calls = 0
move e2e4

how to set nr of cores in scorpio.ini?
here mt is 124 and inside Task Manager Engine is allowed only to use CPU 0 and CPU 1

[Daniel Shawul]

Here is a nice table comparing delay=0 with delay=1

Code: Select all

                 delay=0      delay=1
 1-core           161          16532
 2-core           376          20114
 4-core           21615        23331
 8-core           28756        25208
16-core           29804        25199
32-core           29797        24500

It seems the sweet spot for "delay=0" is 4-cpu cores as its performance jumps by 100x. Note that after 8-cores delay=0 is better than delay=1 whose performance keeps on degrading from then onwards. What delay=1 basically does is insert a Sleep(1) in the threads so that you can have many threads working even on single core.

@Werner. The most important number is the pps/nneps. It seems in your case it is 1957 on command prompt for delay=0 on the initial position. And it seems to vary wildly inside arena (5k for delay=1 and 12k for delay=0 ?) so I am not sure. If you can do the comparison from the command line for 4-cores from the initial position, it would be great. But the safe bet is to use "delay=1" for upto 4 cores so better switch to that I think.

I am working on programatically restricting the number of cpu cores so I will let you know when I am done with it.

regards,
Daniel

[Werner]

From command prompt

delay 0 4CPU
# nodes = 165605 <0% qnodes> time = 11028ms nps = 15016 eps = 0 nneps = 1881
# Tree: nodes = 27987 depth = 19 pps = 1857 visits = 20315
# qsearch_calls = 0 search_calls = 0
move e2e4

delay 1 4CPU
# nodes = 151378 <0% qnodes> time = 11731ms nps = 12904 eps = 0 nneps = 1661
# Tree: nodes = 26307 depth = 18 pps = 1642 visits = 19090
# qsearch_calls = 0 search_calls = 0
move e2e4

and yes the GTX 1650 uses half precision (FP16) with another math model as for rtx Cards and the older Lc0 versions 0.20.0 have had very low nps using FP16.
New Lc0 0.21.2 now has both models implemented for GTX 16... Cards and for RTX 20... Cards.
So Maybe I have to use int8

[Daniel Shawul]

I am not sure what is going on here. Can you post/attach the log file for one of the games if you have it.
I don't trust delay=0 on less than 4-cores even though it seems it is performing slightly better in your case
while in my case it is 100x slower. Btw the 20 sec arena analysis you posted actually run only for 2 and 5 sec for some reason.
I am pretty sure time=xx is processor time ( not wall clock time) so even though 20 sec were spent, scorpio only had access to cpu for 2 and 5 sec !?
delay=1 had higher cpu usage but it is sitll 25% of wall clock time.
Maybe the nps varies wildly in the games when delay=0 depending on cpu usage...

Daniel

[Werner]

Hi Daniel,
I changed to int8 and here is the Output from command prompt:

# nodes = 250742 <0% qnodes> time = 10341ms nps = 24247 eps = 0 nneps = 2980
# Tree: nodes = 41879 depth = 19 pps = 2955 visits = 30428
# qsearch_calls = 0 search_calls = 0
move e2e4

and with delay 0 even better
# nodes = 393006 <0% qnodes> time = 9233ms nps = 42565 eps = 0 nneps = 4985
# Tree: nodes = 62868 depth = 20 pps = 4962 visits = 45587
# qsearch_calls = 0 search_calls = 0
move e2e4

[Daniel Shawul]

More parametric studies.

Using only 1 core and delay=0, the maximum number of threads that gives a "safe" (not the best) nps on my system is 32.

So here delay=0 is actually better even on 1-core

Code: Select all

              delay=0, mt=32            delay=1, mt=32
1-core          10407                          7636
2-core          11528                          8320

With 64 threads launched, you would need atleast 2-cores for delay=0 to perform reasonably

Code: Select all

              delay=0, mt=64            delay=1, mt=64
1-core          342                          16438
2-cores       17707                          15446

The original table with 128 threads requires 4-cores launched for delay=0 to perform reasonably

Code: Select all

              delay=0,mt=128      delay=1,mt=128
 1-core           161          16532
 2-core           376          20114
 4-core           21615        23331
 8-core           28756        25208
16-core           29804        25199
32-core           29797        24500

I need a way to optimize between delay=0/1 and the mt variable for maximizing pps.
Please bear with me Werner, this is not a striaght forward problem to solve since this could be system dependent.

Daniel

[Werner]

Hi Daniel,
I have started the match again now using INT8 which is much faster than HALF on my GTX 1650.

here an explanation from discord LC0 (the GTX 16... Cards have no Tensor cores)

old lc0 can't make use of the non-tensor core fp16 math throughput because it needs NCHW data layout (and right now we use NHWC with fp16 which is needed for tensor cores)

[Daniel Shawul]

Can you post the nps, and did you run the calibration first ?

I would rather you wait a little bit until i figure out what is going on.

Thanks,
Daniel

[Werner]

Ok Daniel,
stopped the match. I did calibration first and then switched to INT8.
Here is the Output with command scorpio.bat go

feature done=0
Number of cores 4 of 4
ht 16777216 X 16 = 256.0 MB
eht 524288 X 8 = 8.0 MB
pht 32768 X 24 = 0.8 MB
treeht 1342169600 X 10 = 12799.9 MB
processors [128]
EgbbProbe 4.3 by Daniel Shawul
egbb_cache 4084 X 8216 = 32.0 MB
180 egbbs loaded !
Loading neural network : c:\Users\6600K\Arena\Engines\ScorpioNet2\net-maddex.uff
nn_cache 131072 X 1552 = 194.0 MB
Loading graph on /gpu:0
0. main_input 7168 = 112 8 8
1. value_head 1 = 1 1 1
2. policy_head 1858 = 1858 1 1
Neural network loaded !
loading_time = 2s
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 11114ms, mt = 29250ms , hply = 0 , moves_left 10]
63 21 117 5210 e2-e3 Ng8-f6 d2-d4 d7-d5 c2-c4 e7-e6 Ng1-f3 Bf8-e7 Bf1-e2
64 28 231 11198 d2-d4 e7-e6 e2-e4 d7-d5 Nb1-c3 Ng8-f6 e4-e5 Nf6-d7 f2-f4 c7-c5 Ng1-f3 Nb8-c6 Bc1-e3 Bf8-e7
65 32 343 17057 e2-e4 c7-c6 d2-d4 d7-d5 Nb1-c3 d5xe4 Nc3xe4 Bc8-f5 Ne4-g3 Bf5-g6 Ng1-f3 Nb8-d7 h2-h4 h7-h6 Bf1-d3
66 35 460 22897 e2-e4 c7-c5 Nb1-c3 Nb8-c6 Ng1-f3 g7-g6 d2-d4 c5xd4 Nf3xd4 Bf8-g7 Bc1-e3 Ng8-f6 Bf1-c4 Ke8-g8 Bc4-b3 d7-d6
67 27 573 29102 d2-d4 Ng8-f6 e2-e3 d7-d5 c2-c4 e7-e6 Ng1-f3 Bf8-e7 Bf1-e2 Ke8-g8 Ke1-g1 d5xc4
68 30 687 35448 e2-e4 c7-c5 Ng1-f3 e7-e6 d2-d4 c5xd4 Nf3xd4 Ng8-f6 Nb1-c3 Nb8-c6 a2-a3 Qd8-c7 Bc1-e3 a7-a6 f2-f4
69 35 804 41273 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-c4 Bf8-c5 Ke1-g1 Ng8-f6 d2-d3 d7-d6 c2-c3 Ke8-g8 Bc1-g5
70 38 892 45048 e2-e4 e7-e5 Ng1-f3 Nb8-c6 d2-d4 e5xd4 Nf3xd4 Ng8-f6 Nb1-c3 Bf8-b4 Nd4xc6 b7xc6 Bf1-d3 d7-d5 e4xd5 c6xd5 Ke1-g1 Ke8-g8 Bc1-g5 c7-c6 Qd1-f3

# Move Value=(V,P,V+P) Policy Visits PV
#----------------------------------------------------------------------------------
# 1 (0.555,0.487,0.521) 20.27 24073 e2-e4 e7-e5 Ng1-f3 Nb8-c6 d2-d4 e5xd4 Nf3xd4 Ng8-f6 Nb1-c3 Bf8-b4 Nd4xc6 b7xc6 Bf1-d3 d7-d5 e4xd5 c6xd5 Ke1-g1 Ke8-g8 Bc1-g5 c7-c6 Qd1-f3
# 2 (0.541,0.487,0.514) 17.81 12446 d2-d4 d7-d5 e2-e3 Ng8-f6 c2-c4 e7-e6 Ng1-f3 Bf8-e7 Bf1-e2 Ke8-g8 Ke1-g1 d5xc4
# 3 (0.524,0.487,0.505) 9.91 2155 Ng1-f3 Ng8-f6 c2-c4 e7-e6 g2-g3 d7-d5 Bf1-g2 Bf8-e7 d2-d4 Ke8-g8 Ke1-g1
# 4 (0.519,0.487,0.503) 9.75 1694 c2-c4 e7-e5 g2-g3 Ng8-f6 Bf1-g2 Bf8-c5 Nb1-c3 Ke8-g8 Ng1-f3 Nb8-c6
# 5 (0.515,0.487,0.501) 5.17 842 e2-e3 Ng8-f6 d2-d4 d7-d5 c2-c4 e7-e6 Ng1-f3 Bf8-e7 Bf1-e2 Ke8-g8
# 6 (0.496,0.487,0.491) 4.85 806 g2-g3 d7-d5 Ng1-f3 c7-c5 Bf1-g2 Ng8-f6 Ke1-g1 Nb8-c6 d2-d4 c5xd4 Nf3xd4 e7-e5
# 7 (0.514,0.487,0.501) 4.08 640 c2-c3 d7-d5 d2-d4 Ng8-f6 Ng1-f3 e7-e6 Bc1-f4 Bf8-d6 e2-e3 Bd6xf4
# 8 (0.500,0.487,0.494) 3.30 348 b2-b3 e7-e5 Bc1-b2 Nb8-c6 e2-e3 Ng8-f6 Ng1-f3 e5-e4 Nf3-d4 Bf8-c5 Nd4xc6 d7xc6
# 9 (0.501,0.487,0.494) 3.27 357 d2-d3 d7-d5 Ng1-f3 Ng8-f6 g2-g3 c7-c5 Bf1-g2 Nb8-c6 Ke1-g1 e7-e5
# 10 (0.518,0.487,0.503) 2.72 507 Nb1-c3 d7-d5 d2-d4 Ng8-f6 Bc1-f4 a7-a6 e2-e3 e7-e6 Ng1-f3 c7-c5 Bf1-e2 Nb8-c6
# 11 (0.480,0.487,0.480) 2.63 191 f2-f4 d7-d5 Ng1-f3 Ng8-f6 e2-e3 c7-c5 b2-b3 g7-g6
# 12 (0.502,0.487,0.495) 2.30 257 h2-h3 d7-d5 d2-d4 c7-c5 e2-e3 Ng8-f6 Ng1-f3 Nb8-c6 c2-c4 e7-e6
# 13 (0.472,0.487,0.472) 2.28 148 b2-b4 e7-e5 Bc1-b2 Bf8xb4 Bb2xe5 Ng8-f6 c2-c3 Bb4-e7 d2-d4 Ke8-g8 e2-e3
# 14 (0.508,0.487,0.498) 1.98 262 a2-a3 d7-d5 d2-d4 Ng8-f6 Ng1-f3 g7-g6 c2-c4 Bf8-g7
# 15 (0.431,0.487,0.431) 1.87 77 Ng1-h3 d7-d5 g2-g3 e7-e5 d2-d4 e5xd4 Qd1xd4 Nb8-c6
# 16 (0.450,0.487,0.450) 1.70 83 h2-h4 d7-d5 d2-d4 c7-c5 d4xc5 Ng8-f6 Ng1-f3 e7-e6
# 17 (0.370,0.487,0.370) 1.64 45 g2-g4 d7-d5 h2-h3 e7-e5 Bf1-g2 Nb8-c6 Nb1-c3 Ng8-e7
# 18 (0.478,0.487,0.478) 1.63 114 a2-a4 e7-e5 e2-e4 Ng8-f6 Nb1-c3 Bf8-b4 Ng1-f3 Ke8-g8
# 19 (0.431,0.487,0.431) 1.53 65 f2-f3 e7-e5 Nb1-c3 Ng8-f6 d2-d4 e5xd4 Qd1xd4
# 20 (0.441,0.487,0.441) 1.32 60 Nb1-a3 e7-e5 e2-e3 d7-d5 d2-d4 Nb8-c6 Ng1-f3 e5-e4

# nodes = 387483 <0% qnodes> time = 9279ms nps = 41759 eps = 0 nneps = 4913
# Tree: nodes = 62282 depth = 20 pps = 4892 visits = 45171
# qsearch_calls = 0 search_calls = 0
move e2e4


and here the files in the new Directory
nnprobe-windows-gpu\
net-maddex.uff.128_2.trt
scorpio.ini
calibrate.dat
speed tests
calibrate.epd
scorpio.bat
scorpio.exe
scorpio.sh
net-maddex.uff