Fat Fritz Net and Leela Nets

[corres]

I run match between Fat Fritz and Leela. I want to know the real chess power of Fat Fritz Net relative to Nets used by Leela. For this reason I modified the parameters of Fat Fritz on the parameters of Leela. It needs because a lot of (chessbase) parameters of Fat Fritz differ from the default parameters of Leela.
The used hardware:
CPU - Threadripper 2950x 16 x 4 GHz
RAM - 32 GB
DUAL GPU - 2 x RTX 2060 OC
OS - Windows 10 64 bits
Parameters of Leela (lc0-v.0.22.0) and Fat Fritz (based on lc0-v.0.22.0)
Threads - 4
Backend - Multiplexing
BackendOptions - backend=cudnn-fp16,(gpu=0),(gpu=1)
NNCacheSize - 10000000
MaxBatchPrefetch - 64
Others are as leela`s Default
Endgame database - 6 men syzygy
Openings - 50 positions with alternated colors (my TestBook50)
Time Control - 1 min + 2 sec / move

Result
Net of Fat Fritz : 42872 (Leela Net) = 6 : 3 (91 draw) 100 games

As my earlier tests proved at least on my system the most powerful net of Leela is the 42872.
This test shows the chess power of Fat Fritz net may be higher in some measure.
Comparing this result to others the original ChassBase parameters of Fat Fritz may be not optimal.

[dkappe]

I think various cpuct settings and policy softmax temp are also different.

[corres]

I think various cpuct settings and policy softmax temp are also different.

I EVERY parameters of Fat Fritz modified as they are in Leela.

[dkappe]

I think various cpuct settings and policy softmax temp are also different.

I EVERY parameters of Fat Fritz modified as they are in Leela.

I don’t understand. You did or did not change the cpuctbase, cpuctfactor, policy softmax temp to match those in the Fritz 17 executable?

[corres]

I think various cpuct settings and policy softmax temp are also different.

I EVERY parameters of Fat Fritz modified as they are in Leela.

I don’t understand. You did or did not change the cpuctbase, cpuctfactor, policy softmax temp to match those in the Fritz 17 executable?

I use DeepFritz 14 GUI and not use config file of Fat Fritz.
But you can also modify in Fritz 17 GUI the parameters of Fat Fritz.

[shrapnel]

This test shows the chess power of Fat Fritz net may be higher in some measure.

Really ?? Watch this Game, boyz...
[pgn][Event "Rated game, 5m+5s"]
[Site "Engine Room"]
[Date "2019.11.24"]
[Round "?"]
[White "Shrapnel, Lc0 v0.22.0"]
[Black "Pestif, Fat Fritz (in lc0) "]
[Result "1-0"]
[ECO "C92"]
[WhiteElo "2466"]
[BlackElo "2295"]
[Annotator "0.28;0.47"]
[PlyCount "90"]
[EventDate "2019.11.24"]
[SourceTitle "playchess.com"]
[TimeControl "300+5"]

{Fat Fritz (in lc0) cpu (12 threads): 9.7 plies; 0kN/s AMD Ryzen 7 2700X
Eight-Core Processor 3700MHz, (8 cores, 16 threads), 52356 MB} 1. e4 {
0.28/15 13} e5 {B 0} 2. Nf3 {0.26/18 6} Nc6 {B 0} 3. Bb5 {0.26/20 7} a6 {
B 0 (Nf6)} 4. Ba4 {0.46/16 10} Nf6 {B 0} 5. O-O {0.44/17 6} Be7 {B 0 (b5)} 6.
Re1 {0.48/18 11} b5 {B 0} 7. Bb3 {0.48/18 4} O-O {B 0 (d6)} 8. c3 {0.48/17 14}
d6 {B 0} 9. h3 {0.48/16 2} Re8 {B 0 (Bb7)} 10. d4 {0.47/16 23} Bb7 {0.47/6 12}
11. Nbd2 {0.38/19 28} Bf8 {B 0} 12. a4 {0.38/18 0} exd4 {0.47/10 14 (Na5)} 13.
cxd4 {0.49/15 9} h6 {0.49/8 11 (Qd7)} 14. Bc2 {0.48/14 24 (Ba2)} Nb4 {0.53/10
11} 15. Bb1 {0.36/21 0} c5 {0.54/11 11} 16. d5 {0.31/16 0} Nd7 {0.54/12 11} 17.
Ra3 {0.29/25 0} c4 {0.49/12 10} 18. Nd4 {0.26/26 6} Ne5 {0.37/12 25 (Qf6)} 19.
axb5 {0.52/17 20} Nbd3 {0.40/12 8} 20. Re3 {0.45/18 9 (Bxd3)} axb5 {0.31/9 18
(Nxc1)} 21. Rxa8 {0.58/20 15} Qxa8 {0.26/9 8} 22. Nxb5 {0.55/20 1} Qa1 {
0.26/9 18} 23. f4 {0.56/20 1 (b3)} Nxc1 {0.04/8 14 (Qa5)} 24. Qxc1 {0.77/19 24}
Nd3 {0.49/9 8} 25. Qxc4 {0.79/20 4} Nxf4 {0.91/10 12} 26. Qa2 {0.79/21 0} Qxa2
{0.98/9 11} 27. Bxa2 {0.77/20 1} g5 {0.95/7 30 (f5)} 28. Ra3 {1.31/18 31} f5 {
0.80/8 13} 29. Ra7 {1.34/19 1} Rb8 {1.19/8 12 (Bc8)} 30. Nd4 {2.21/15 25} Bc8 {
1.59/8 11} 31. b3 {2.31/15 9} fxe4 {1.64/9 23 (Kh8)} 32. Nxe4 {2.47/18 17} Rb7
{1.53/10 5} 33. Ra8 {2.55/21 5 (Rxb7)} Re7 {1.57/8 14} 34. Rxc8 {2.73/20 4}
Rxe4 {2.19/9 10} 35. Ne6 {2.86/18 2} Nxe6 {1.74/10 10} 36. dxe6 {2.95/17 3} d5
{1.77/10 11} 37. b4 {3.09/18 6} Kg7 {1.94/9 6} 38. Rc7+ {3.23/17 4} Kh8 {
1.82/10 12} 39. Bxd5 {3.32/16 4} Re5 {2.01/8 13 (Re1+)} 40. b5 {4.20/12 28}
Rxd5 {2.24/9 8} 41. b6 {4.98/11 18} Rb5 {2.79/7 26} 42. b7 {6.07/10 13} Rb1+ {
2.81/7 23 (Bc5+)} 43. Kf2 {6.21/9 14} Rb2+ {3.04/6 13} 44. Kf3 {6.90/9 24} Kg8
{4.07/5 11 (g4+)} 45. e7 {11.90/7 13} Bxe7 {10.28/4 11 Pestif,Fat Fritz (in
lc0) c resigns (Lag: Av=0.22s, max=0.4s)} 1-0[/pgn]

[Damir]

CPU version of Fat Fritz with 0 kN/s vs your 3 GPU's… something to brag about. Come again some other time.

[corres]

This test shows the chess power of Fat Fritz net may be higher in some measure.

Really ?? Watch this Game, boyz...
[pgn][Event "Rated game, 5m+5s"]
[Site "Engine Room"]
[Date "2019.11.24"]
[Round "?"]
[White "Shrapnel, Lc0 v0.22.0"]
[Black "Pestif, Fat Fritz (in lc0) "]
[Result "1-0"]
[ECO "C92"]
[WhiteElo "2466"]
[BlackElo "2295"]
[Annotator "0.28;0.47"]
[PlyCount "90"]
[EventDate "2019.11.24"]
[SourceTitle "playchess.com"]
[TimeControl "300+5"]

{Fat Fritz (in lc0) cpu (12 threads): 9.7 plies; 0kN/s AMD Ryzen 7 2700X
Eight-Core Processor 3700MHz, (8 cores, 16 threads), 52356 MB} 1. e4 {
0.28/15 13} e5 {B 0} 2. Nf3 {0.26/18 6} Nc6 {B 0} 3. Bb5 {0.26/20 7} a6 {
B 0 (Nf6)} 4. Ba4 {0.46/16 10} Nf6 {B 0} 5. O-O {0.44/17 6} Be7 {B 0 (b5)} 6.
Re1 {0.48/18 11} b5 {B 0} 7. Bb3 {0.48/18 4} O-O {B 0 (d6)} 8. c3 {0.48/17 14}
d6 {B 0} 9. h3 {0.48/16 2} Re8 {B 0 (Bb7)} 10. d4 {0.47/16 23} Bb7 {0.47/6 12}
11. Nbd2 {0.38/19 28} Bf8 {B 0} 12. a4 {0.38/18 0} exd4 {0.47/10 14 (Na5)} 13.
cxd4 {0.49/15 9} h6 {0.49/8 11 (Qd7)} 14. Bc2 {0.48/14 24 (Ba2)} Nb4 {0.53/10
11} 15. Bb1 {0.36/21 0} c5 {0.54/11 11} 16. d5 {0.31/16 0} Nd7 {0.54/12 11} 17.
Ra3 {0.29/25 0} c4 {0.49/12 10} 18. Nd4 {0.26/26 6} Ne5 {0.37/12 25 (Qf6)} 19.
axb5 {0.52/17 20} Nbd3 {0.40/12 8} 20. Re3 {0.45/18 9 (Bxd3)} axb5 {0.31/9 18
(Nxc1)} 21. Rxa8 {0.58/20 15} Qxa8 {0.26/9 8} 22. Nxb5 {0.55/20 1} Qa1 {
0.26/9 18} 23. f4 {0.56/20 1 (b3)} Nxc1 {0.04/8 14 (Qa5)} 24. Qxc1 {0.77/19 24}
Nd3 {0.49/9 8} 25. Qxc4 {0.79/20 4} Nxf4 {0.91/10 12} 26. Qa2 {0.79/21 0} Qxa2
{0.98/9 11} 27. Bxa2 {0.77/20 1} g5 {0.95/7 30 (f5)} 28. Ra3 {1.31/18 31} f5 {
0.80/8 13} 29. Ra7 {1.34/19 1} Rb8 {1.19/8 12 (Bc8)} 30. Nd4 {2.21/15 25} Bc8 {
1.59/8 11} 31. b3 {2.31/15 9} fxe4 {1.64/9 23 (Kh8)} 32. Nxe4 {2.47/18 17} Rb7
{1.53/10 5} 33. Ra8 {2.55/21 5 (Rxb7)} Re7 {1.57/8 14} 34. Rxc8 {2.73/20 4}
Rxe4 {2.19/9 10} 35. Ne6 {2.86/18 2} Nxe6 {1.74/10 10} 36. dxe6 {2.95/17 3} d5
{1.77/10 11} 37. b4 {3.09/18 6} Kg7 {1.94/9 6} 38. Rc7+ {3.23/17 4} Kh8 {
1.82/10 12} 39. Bxd5 {3.32/16 4} Re5 {2.01/8 13 (Re1+)} 40. b5 {4.20/12 28}
Rxd5 {2.24/9 8} 41. b6 {4.98/11 18} Rb5 {2.79/7 26} 42. b7 {6.07/10 13} Rb1+ {
2.81/7 23 (Bc5+)} 43. Kf2 {6.21/9 14} Rb2+ {3.04/6 13} 44. Kf3 {6.90/9 24} Kg8
{4.07/5 11 (g4+)} 45. e7 {11.90/7 13} Bxe7 {10.28/4 11 Pestif,Fat Fritz (in
lc0) c resigns (Lag: Av=0.22s, max=0.4s)} 1-0[/pgn]

Instead of non-equalized games you ought to show us tests with your twin RTX 2080 Ti.
Why do you grudge the time and money from tests?
One test spends one day if you use my TestBook50 and Time Control.
We can wait you to make tests helping the community of talkchess.
Do you not think in this way?

[shrapnel]

CPU version of Fat Fritz with 0 kN/s vs your 3 GPU's… something to brag about. Come again some other time.

I told you before, I can't help it if the opponent uses weak hardware.
Why don't you try me on Playchess ?
I'll only be too happy to give you a Beating.

[shrapnel]

Instead of non-equalized games you ought to show us tests with your twin RTX 2080 Ti.
Why do you grudge the time and money from tests?
One test spends one day if you use my TestBook50 and Time Control.
We can wait you to make tests helping the community of talkchess.
Do you not think in this way?

No, I don't think in this way.

Neither am I interested in nor do I have the Time for Tests.
I do all my Testing through actual games with other Players, weak and strong.
Chess is a Game and I'm a Gamer, not a Tester.
I'll leave the boring Testing to guys like you.
If you want me try out some new settings in the the Real Gaming World, just drop me a word.
Regards