Testing other Networks: 58573 and LS 13.0
2 verfasser
Seite 1 von 1
Testing other Networks: 58573 and LS 13.0
The tests with 58573 did not get much better performance : 3457 - very close to LCZero 0.21.2 Cuda (42668) with 3455.
I will now test the LS 13.0 net with LCZero 0.23.2 as the Leelenstein Engine compile from December 26th does not work with fp16 on my GTX 1650 cards.
1 LCZero 0.23.2 Cuda (LS13.0) +28 +14/=80/-6 54.00% 54.0/100
2 Fat Fritz 1.0 (in lc0) Cuda -28 +6/=80/-14 46.00% 46.0/100
1 LCZero 0.23.2 Cuda (LS13.0) 3513 +269 +68/=29/-3 82.50% 82.5/100
2 Ethereal 11.75 x64 1CPU 3244 -269 +3/=29/-68 17.50% 17.5/100
repetition with ratio 1.3:
1 LCZero 0.23.1 Cuda (LS13.0) +225 = 3469 +61/=35/-4 78.50% 78.5/100
2 Ethereal 11.75 x64 1CPU -225 +4/=35/-61 21.50% 21.5/100
1 LCZero 0.23.2 Cuda (LS13.0) 2200 +34 +22/=66/-12 55.00% 55.0/100
2 Stockfish 10.0 x64 1CPU 2200 -34 +12/=66/-22 45.00% 45.0/100
1 LCZero 0.23.1 Cuda (LS13.0) +127 +42/=51/-7 67.50% 67.5/100
2 Stoofvlees II a13 -127 +7/=51/-42 32.50% 32.5/100
1 LCZero 0.23.1 Cuda (LS13.0) +173 +51/=44/-5 73.00% 73.0/100
2 Komodo 13.2.5 x64 1CPU -173 +5/=44/-51 27.00% 27.0/100
1 LCZero 0.23.1 Cuda (LS13.0) 2200 +139 +44/=53/-5 69.12% 70.5/102
2 Stoofvlees II a12 2200 -139 +5/=53/-44 30.88% 31.5/102
LCZero 0.23.1 Cuda (LS13.0) - Stockfish 11.0 x64 1CPU 51.5 - 48.5 +15/=73/-12 51.50%
LCZero 0.23.1 Cuda (LS13.0) - Komodo 13.2.5 x64 1CPU (MCTS) 79.5 - 20.5 +61/=37/-2 79.50%
1 LCZero 0.23.1 Cuda (LS13.0) +35 = 3384 +34/=42/-24 55.00% 55.0/100
2 Komodo 13.2.5 x64 1CPU -35 +24/=42/-34 45.00% 45.0/100
and results with 1CPU (Blas) and 58573 net
Lc0 0.23.2a 58573 - SlowChess Blitz Classic 1.7 53.5 - 46.5 +28/=51/-21 53.50%
Lc0 0.23.2a 58573 - Demolito 2019-12-29 x64 1CPU 57.5 - 42.5 +34/=47/-19 57.50%
Lc0 0.23.2a 58573 - Arasan 22.0 x64 1CPU 43.0 - 57.0 +19/=48/-33 43.00
1 Lc0 0.23.2a 58573 2200 +147 +50/=40/-10 70.00% 70.0/100
2 PeSTO 2.210 x64 1CPU 2200 -147 +10/=40/-50 30.00% 30.0/100
1 Pedone 2.0 x64 1CPU 2200 +6 +31/=40/-29 51.00% 51.0/100
2 Lc0 CPU 0.23.2a 58573 2200 -6 +29/=40/-31 49.00% 49.0/100
1 Chess22k 1.13 x64 2200 +3 +29/=43/-28 50.50% 50.5/100
2 Lc0 CPU 0.23.2a 58573 2200 -3 +28/=43/-29 49.50% 49.5/100
1 LCZero 0.23.2 Blas (58573) 3044 +35 +35/=40/-25 55.00% 55.0/100
2 Vajolet2 2.8 x64 1CPU 3009 -35 +25/=40/-35 45.00% 45.0/100
1 Critter 1.6 x64 1CPU 3012 +104 +42/=45/-13 64.50% 64.5/100
2 LCZero 0.23.2 Blas (58573) 2908 -104 +13/=45/-42 35.50% 35.5/100
1 SlowChess Blitz Classic 1.8 2990? +42 +30/=52/-18 56.00% 56.0/100
2 LCZero 0.23.2 Blas (58573) 2948? -42 +18/=52/-30 44.00% 44.0/100
I will now test the LS 13.0 net with LCZero 0.23.2 as the Leelenstein Engine compile from December 26th does not work with fp16 on my GTX 1650 cards.
1 LCZero 0.23.2 Cuda (LS13.0) +28 +14/=80/-6 54.00% 54.0/100
2 Fat Fritz 1.0 (in lc0) Cuda -28 +6/=80/-14 46.00% 46.0/100
1 LCZero 0.23.2 Cuda (LS13.0) 3513 +269 +68/=29/-3 82.50% 82.5/100
2 Ethereal 11.75 x64 1CPU 3244 -269 +3/=29/-68 17.50% 17.5/100
repetition with ratio 1.3:
1 LCZero 0.23.1 Cuda (LS13.0) +225 = 3469 +61/=35/-4 78.50% 78.5/100
2 Ethereal 11.75 x64 1CPU -225 +4/=35/-61 21.50% 21.5/100
1 LCZero 0.23.2 Cuda (LS13.0) 2200 +34 +22/=66/-12 55.00% 55.0/100
2 Stockfish 10.0 x64 1CPU 2200 -34 +12/=66/-22 45.00% 45.0/100
1 LCZero 0.23.1 Cuda (LS13.0) +127 +42/=51/-7 67.50% 67.5/100
2 Stoofvlees II a13 -127 +7/=51/-42 32.50% 32.5/100
1 LCZero 0.23.1 Cuda (LS13.0) +173 +51/=44/-5 73.00% 73.0/100
2 Komodo 13.2.5 x64 1CPU -173 +5/=44/-51 27.00% 27.0/100
1 LCZero 0.23.1 Cuda (LS13.0) 2200 +139 +44/=53/-5 69.12% 70.5/102
2 Stoofvlees II a12 2200 -139 +5/=53/-44 30.88% 31.5/102
LCZero 0.23.1 Cuda (LS13.0) - Stockfish 11.0 x64 1CPU 51.5 - 48.5 +15/=73/-12 51.50%
LCZero 0.23.1 Cuda (LS13.0) - Komodo 13.2.5 x64 1CPU (MCTS) 79.5 - 20.5 +61/=37/-2 79.50%
1 LCZero 0.23.1 Cuda (LS13.0) +35 = 3384 +34/=42/-24 55.00% 55.0/100
2 Komodo 13.2.5 x64 1CPU -35 +24/=42/-34 45.00% 45.0/100
and results with 1CPU (Blas) and 58573 net
Lc0 0.23.2a 58573 - SlowChess Blitz Classic 1.7 53.5 - 46.5 +28/=51/-21 53.50%
Lc0 0.23.2a 58573 - Demolito 2019-12-29 x64 1CPU 57.5 - 42.5 +34/=47/-19 57.50%
Lc0 0.23.2a 58573 - Arasan 22.0 x64 1CPU 43.0 - 57.0 +19/=48/-33 43.00
1 Lc0 0.23.2a 58573 2200 +147 +50/=40/-10 70.00% 70.0/100
2 PeSTO 2.210 x64 1CPU 2200 -147 +10/=40/-50 30.00% 30.0/100
1 Pedone 2.0 x64 1CPU 2200 +6 +31/=40/-29 51.00% 51.0/100
2 Lc0 CPU 0.23.2a 58573 2200 -6 +29/=40/-31 49.00% 49.0/100
1 Chess22k 1.13 x64 2200 +3 +29/=43/-28 50.50% 50.5/100
2 Lc0 CPU 0.23.2a 58573 2200 -3 +28/=43/-29 49.50% 49.5/100
1 LCZero 0.23.2 Blas (58573) 3044 +35 +35/=40/-25 55.00% 55.0/100
2 Vajolet2 2.8 x64 1CPU 3009 -35 +25/=40/-35 45.00% 45.0/100
1 Critter 1.6 x64 1CPU 3012 +104 +42/=45/-13 64.50% 64.5/100
2 LCZero 0.23.2 Blas (58573) 2908 -104 +13/=45/-42 35.50% 35.5/100
1 SlowChess Blitz Classic 1.8 2990? +42 +30/=52/-18 56.00% 56.0/100
2 LCZero 0.23.2 Blas (58573) 2948? -42 +18/=52/-30 44.00% 44.0/100
Zuletzt von Werner am Sa Feb 01, 2020 11:59 am bearbeitet; insgesamt 13-mal bearbeitet
Re: Testing other Networks: 58573 and LS 13.0
I'll add some games for Leelenstein 13.0
50 games each (added to existing matches)
100 games each (new matches)
Hardware
GPU: GTX 1060 6GB
CPU: Intel i7-4770 @ 3,4 GHZ (1Core)
Ratio: ca. 1,3
50 games each (added to existing matches)
vs. Komodo 13.2.5 x64 1CPU 35,0:15,0 perf=3493
vs. Ethereal 11.75 x64 1CPU 36,0:14,0 perf=3406
vs. Stoofvlees II a12
100 games each (new matches)
vs. Xiphos 0.6 x64 1CPU 76,5:23,5 perf=3426
vs. Fire 7.1 x64 1CPU
vs. Komodo 13.2.5 x64 1CPU MCTS
Hardware
GPU: GTX 1060 6GB
CPU: Intel i7-4770 @ 3,4 GHZ (1Core)
Ratio: ca. 1,3
Wolfgang- Admin
- Anzahl der Beiträge : 1606
Anmeldedatum : 23.10.13
Ähnliche Themen
» Testing Mr Bob 1.0/1.1/1.2
» Testing some more (small) LCZero Networks 128x10
» testing K2 0.87
» testing Isa 2.0.48
» Testing SlowChessClassic 2.2 x64
» Testing some more (small) LCZero Networks 128x10
» testing K2 0.87
» testing Isa 2.0.48
» Testing SlowChessClassic 2.2 x64
Seite 1 von 1
Befugnisse in diesem Forum
Sie können in diesem Forum nicht antworten
|
|