Lc0 in GTX video card

[Ferdy]

I tested the 2 versions on my GTX 1650 Super under the same conditions with go movetime 60000; NN 10x128

LC0 Cuda 0.26.3 backend cuda nps 19247
LC0 Cuda 0.26.3 backend cuda-fp16 nps 6634

LC0 CUDNN 0.26.3 backend cudnn nps 22990
LC0 CUDNN 0.26.3 backend cudnn-fp16 nps 37701

LC0 CUDNN 0.26.3 backend cuda nps 23732
LC0 CUDNN 0.26.3 backend cuda-fp16 nps 37573

So I think, the older cudnn version maybe the strongest with this card.

What do you mean by "the older cudnn version"?

[Werner]

Sorry ,
the cuda drivers are older only, the lc0 version is the same:
cublasLt64_11.dll 214.235.648 16.09.2020 05:17 -a-- in the CUDA package
or
cublas64_100.dll 67.318.784 26.08.2018 06:33 -a-- in the CUDNN package

clear now?

[Jouni]

I my new test even 1cpu SF12 beats Lc0 in GTX card . Time to order A100 or two..

[Ferdy]

I my new test even 1cpu SF12 beats Lc0 in GTX card . Time to order A100 or two..

Classic loses to GTX card, but Lc0 likes to do something else . Are Lc0 developers cannot solve this issue?

[pgn][Event "Computer Games"]
[Site "CuteChess GUI"]
[Date "2021.01.01"]
[Round "1"]
[White "Stockfish 12 Classic"]
[Black "Lc0 v0.26.3 w42850 20x256"]
[Result "1/2-1/2"]
[ECO "A41"]
[GameDuration "00:07:06"]
[GameEndTime "2021-01-01T07:00:05.280 Taipei Standard Time"]
[GameStartTime "2021-01-01T06:52:58.433 Taipei Standard Time"]
[Opening "Old Indian defense"]
[TimeControl "60+1"]

1.d4 { book } 1...d6 { book } 2.c4 { book } 2...e5 { book } 3.d5 { book } 3...g6 { book } 4.e4 { +0.98/24 3.98s } 4...Nd7 { -0.18/8 1.48s } 5.Nc3 { +0.92/25 1.39s } 5...a5 { -0.15/9 1.89s } 6.Bd3 { +1.00/25 2.16s } 6...h5 { -0.14/10 2.95s } 7.Nge2 { +1.16/21 0.93s } 7...Bh6 { -0.13/8 1.73s } 8.O-O { +1.25/22 1.32s } 8...Bxc1 { -0.09/8 1.94s } 9.Nxc1 { +0.89/28 14.19s } 9...Ngf6 { -0.08/8 3.27s } 10.Nb3 { +0.86/24 4.59s } 10...h4 { 0.00/8 2.46s } 11.Be2 { +0.71/23 1.02s } 11...b6 { -0.04/8 1.86s } 12.h3 { +0.94/23 1.35s } 12...Nh7 { -0.01/7 2.40s } 13.Qd3 { +0.60/26 11.56s } 13...Qg5 { -0.02/8 1.60s } 14.Rad1 { +0.96/21 1.44s } 14...Nhf6 { +0.01/8 2.46s } 15.Kh1 { +0.32/25 6.94s } 15...Qf4 { +0.26/8 1.63s } 16.Bf3 { +0.12/21 1.11s } 16...Nc5 { +0.26/8 1.74s } 17.Nxc5 { 0.00/24 2.74s } 17...bxc5 { +0.26/8 1.15s } 18.Rde1 { -0.13/26 2.18s } 18...Bd7 { +0.24/8 4.05s } 19.a4 { -0.12/21 0.93s } 19...Rb8 { +0.38/7 2.96s } 20.Qc2 { -0.12/25 1.96s } 20...Kd8 { +0.40/8 1.65s } 21.Re3 { +0.31/22 1.59s } 21...Rg8 { +0.49/7 1.39s } 22.Kg1 { -0.12/26 7.40s } 22...Qh6 { +0.58/8 2.13s } 23.Qe2 { -0.40/23 3.07s } 23...Nh7 { +0.53/8 1.97s } 24.Nb5 { +0.17/20 2.17s } 24...Ra8 { +0.58/9 2.69s } 25.Ra3 { +0.16/20 2.55s } 25...f5 { +0.66/8 3.74s } 26.Re1 { 0.00/23 2.13s } 26...Nf6 { +0.63/6 2.30s } 27.Nc3 { 0.00/25 0.75s } 27...Qf4 { +0.59/6 2.37s } 28.Qc2 { 0.00/23 0.32s } 28...Rb8 { +0.84/9 1.13s } 29.Re3 { -0.62/23 4.06s } 29...Rb4 { +1.16/10 1.59s } 30.b3 { -0.63/18 0.29s } 30...Qh6 { +1.07/7 4.03s } 31.Re1 { -0.77/20 1.55s } 31...g5 { +1.27/9 1.58s } 32.exf5 { -0.87/21 0.42s } 32...g4 { +1.22/12 0.99s } 33.hxg4 { -1.17/23 1.09s } 33...Nxg4 { +1.24/11 1.72s } 34.Bxg4 { -1.27/23 1.26s } 34...Rxg4 { +1.15/10 1.10s } 35.Nd1 { -1.17/20 0.31s } 35...Rf4 { +1.16/9 2.99s } 36.Ne3 { -1.57/20 1.07s } 36...Kc8 { +1.16/9 2.38s } 37.Raa1 { -1.63/21 1.53s } 37...Kb7 { +1.46/7 2.45s } 38.Kh2 { -2.00/21 0.54s } 38...Ka6 { +1.56/8 2.49s } 39.Rh1 { -2.01/24 1.35s } 39...Qg5 { +1.66/10 2.97s } 40.Kg1 { -1.94/23 0.68s } 40...Rb8 { +1.76/10 0.66s } 41.Re1 { -2.11/22 0.74s } 41...Rg8 { +1.96/11 1.74s } 42.Rh2 { -1.75/20 0.36s } 42...Bxf5 { +2.17/10 1.53s } 43.Nxf5 { -2.52/24 3.97s } 43...Rxf5 { +2.19/9 0.98s } 44.Qe4 { -2.52/21 0.63s } 44...Rf4 { +2.60/9 2.76s } 45.Qe2 { -2.67/23 1.57s } 45...Qf5 { +3.47/7 3.64s } 46.Rh1 { -2.58/19 0.14s } 46...h3 { +3.74/7 3.03s } 47.g3 { -3.09/23 0.81s } 47...Rf8 { +4.39/9 2.02s } 48.Rh2 { -2.60/20 0.43s } 48...Rf3 { +4.49/8 0.85s } 49.Qf1 { -3.01/24 0.57s } 49...Rh8 { +4.61/9 2.02s } 50.Qe2 { -3.18/26 0.73s } 50...Rxb3 { +14.08/6 1.47s } 51.Rh1 { -2.94/23 0.32s } 51...Rf8 { +11.12/7 2.84s } 52.Kh2 { -3.40/26 2.27s } 52...Rfb8 { +16.02/7 1.13s } 53.Rd1 { -3.45/23 0.49s } 53...R8b4 { +11.80/6 1.91s } 54.Rd2 { -3.02/19 0.34s } 54...Rxa4 { +8.14/6 1.47s } 55.Rf1 { -4.10/23 3.75s } 55...Qf3 { +8.30/7 0.96s } 56.Qxf3 { -3.40/20 0.34s } 56...Rxf3 { +8.10/8 0.69s } 57.Rc2 { -3.40/20 0.31s } 57...Rb4 { +8.24/8 1.74s } 58.Kxh3 { -3.44/21 0.21s } 58...a4 { +8.86/8 0.66s } 59.Kg4 { -3.28/24 1.52s } 59...Rbb3 { +10.71/7 1.08s } 60.Rh1 { -3.59/25 0.50s } 60...a3 { +10.78/7 1.97s } 61.Rh8 { -3.78/24 0.68s } 61...Kb7 { +11.02/6 1.53s } 62.Rh1 { -3.71/24 0.71s } 62...Kb6 { +8.66/6 1.65s } 63.Rh8 { -3.25/24 0.36s } 63...Rfc3 { +13.72/6 0.60s } 64.Rxc3 { -2.11/21 0.47s } 64...Rxc3 { +10.33/7 0.82s } 65.Ra8 { -2.97/22 1.63s } 65...Rxc4+ { +10.04/8 0.84s } 66.f4 { -2.72/22 0.45s } 66...Rd4 { +9.15/8 1.59s } 67.Rxa3 { -3.26/23 1.39s } 67...c4 { +9.01/7 0.73s } 68.Ra1 { -3.11/23 3.15s } 68...Kc5 { +10.47/7 1.31s } 69.Kg5 { -3.06/22 0.88s } 69...Rd3 { +13.63/7 1.74s } 70.g4 { -3.18/20 0.52s } 70...exf4 { +11.78/7 0.61s } 71.Kxf4 { -2.75/19 0.35s } 71...Rd4+ { +10.16/7 1.02s } 72.Kf3 { -4.01/22 1.76s } 72...Rxd5 { +9.37/7 1.38s } 73.g5 { -4.33/23 1.34s } 73...Rf5+ { +10.25/7 0.60s } 74.Ke3 { -4.41/24 0.43s } 74...Re5+ { +11.56/7 1.42s } 75.Kf4 { -5.44/22 3.03s } 75...Re8 { +17.33/7 0.91s } 76.Ra5+ { -5.32/20 0.43s } 76...Kd4 { +47.60/5 1.42s } 77.Ra6 { -4.08/20 0.45s } 77...c3 { +48.70/5 1.07s } 78.g6 { -6.60/20 1.71s } 78...c2 { +46.34/5 1.03s } 79.Ra1 { -10.59/18 1.85s } 79...Kc3 { +50.27/5 0.99s } 80.Kg3 { -21.02/19 1.05s } 80...Rg8 { +67.87/4 1.06s } 81.Kf4 { -59.91/27 0.60s } 81...Rxg6 { +66.70/4 1.01s } 82.Kf5 { -65.67/24 0.83s } 82...Rg2 { +71.57/4 1.01s } 83.Rc1 { -72.85/26 1.59s } 83...d5 { +78.46/4 0.99s } 84.Ke6 { -62.29/21 0.17s } 84...d4 { +77.78/4 1.03s } 85.Re1 { -72.76/19 1.78s } 85...d3 { +74.93/4 1.01s } 86.Ra1 { -M24/19 0.91s } 86...Re2+ { +72.98/4 0.99s } 87.Kf7 { -M24/20 0.53s } 87...Re4 { +62.78/4 1.01s } 88.Kg6 { -74.02/19 1.16s } 88...d2 { +77.10/4 1.02s } 89.Kf5 { -M20/25 0.36s } 89...Re3 { +73.73/4 0.99s } 90.Ra3+ { -M20/22 0.24s } 90...Kb4 { +52.31/5 1.01s } 91.Rxe3 { -M18/29 0.25s } 91...c1=Q { +67.15/4 1.00s } 92.Re4+ { -M22/24 0.33s } 92...Kc5 { +36.29/5 0.99s } 93.Re5+ { -M16/27 0.88s } 93...Kc4 { +48.41/5 1.00s } 94.Re4+ { -M18/31 0.27s } 94...Kb3 { +36.84/5 0.96s } 95.Rd4 { -M20/37 0.42s } 95...Qe1 { +49.94/4 1.05s } 96.Rd3+ { -M24/28 0.52s } 96...Kc2 { +64.90/5 1.00s } 97.Rd4 { -M20/40 0.84s } 97...Qg1 { +57.46/4 0.99s } 98.Rd7 { -M22/28 0.55s } 98...Qb1 { +37.49/4 1.01s } 99.Kf6 { -M24/25 0.63s } 99...Qc1 { +26.91/4 1.01s } 100.Rxc7+ { -59.90/26 0.58s } 100...Kb3 { +48.94/5 1.01s } 101.Rb7+ { -M24/43 3.12s } 101...Ka4 { +51.59/5 0.99s } 102.Ra7+ { -M20/44 1.17s } 102...Kb5 { +42.32/5 0.98s } 103.Rd7 { -M18/51 0.46s } 103...Kb6 { +35.66/4 1.00s } 104.Ke5 { -M22/35 0.48s } 104...Qe1+ { +44.26/4 0.95s } 105.Kf4 { -M20/46 0.75s } 105...Qg1 { +50.48/5 1.06s } 106.Rxd2 { -5.26/20 0.56s } 106...Qc1 { +46.79/5 0.88s } 107.Ke3 { -5.10/20 0.66s } 107...Kb5 { +49.96/5 1.10s } 108.Kd3 { -5.04/24 0.88s } 108...Kb4 { +44.39/5 0.98s } 109.Rf2 { -6.10/27 2.88s } 109...Qd1+ { +33.52/5 1.05s } 110.Ke3 { -6.30/31 0.93s } 110...Kc4 { +34.02/5 0.99s } 111.Rf4+ { -5.39/19 0.74s } 111...Kc3 { +38.02/5 0.94s } 112.Ke4 { -4.93/27 0.53s } 112...Qh5 { +36.73/5 1.08s } 113.Rf6 { -5.43/33 2.44s } 113...Kc4 { +28.49/5 0.98s } 114.Rc6+ { -5.43/29 0.39s } 114...Kb5 { +33.69/5 0.93s } 115.Rf6 { -5.43/31 1.17s } 115...Qh1+ { +44.83/5 1.06s } 116.Ke5 { -5.43/31 1.00s } 116...Qh2+ { +30.79/5 1.02s } 117.Ke6 { -5.43/33 0.35s } 117...Qh3+ { +39.98/5 1.01s } 118.Kf7 { -5.60/30 3.62s } 118...Kc5 { +40.98/5 0.98s } 119.Kg7 { -6.63/25 2.26s } 119...Kd5 { +39.23/5 0.99s } 120.Rh6 { -17.73/21 0.65s } 120...Qd7+ { +42.26/5 0.99s } 121.Kf8 { -57.85/24 2.35s } 121...Qg4 { +39.00/5 1.03s } 122.Rf6 { -5.60/21 0.26s } 122...Ke5 { +28.78/5 0.99s } 123.Rf7 { -60.22/22 0.66s } 123...Ke6 { +22.96/5 0.98s } 124.Re7+ { -57.87/22 0.23s } 124...Kd6 { +25.31/6 0.98s } 125.Rh7 { -M42/29 2.10s } 125...Qf3+ { +27.50/5 1.05s } 126.Kg7 { -M42/30 0.38s } 126...Ke6 { +32.33/5 0.95s } 127.Rh6+ { -M38/35 0.24s } 127...Ke7 { +21.85/5 1.05s } 128.Rg6 { -M34/36 0.34s } 128...Qf7+ { +19.76/5 1.00s } 129.Kh6 { -6.30/21 0.42s } 129...Qc4 { +22.47/5 1.02s } 130.Kh5 { -34.91/23 1.28s } 130...Qc1 { +30.90/5 0.99s } 131.Rg3 { -6.71/23 0.70s } 131...Qd1+ { +23.88/5 0.98s } 132.Kg5 { -10.67/25 1.41s } 132...Qd2+ { +26.56/5 0.96s } 133.Kf5 { -13.70/24 1.23s } 133...Qf2+ { +21.71/5 0.98s } 134.Kg4 { -27.05/24 1.20s } 134...Kd6 { +19.31/5 1.07s } 135.Ra3 { -23.09/25 2.29s } 135...Kd5 { +25.10/5 0.99s } 136.Ra8 { -11.49/17 0.40s } 136...Kc4 { +27.42/5 0.98s } 137.Ra6 { -12.22/21 1.49s } 137...Qg2+ { +22.69/5 1.03s } 138.Kf5 { -5.60/22 0.44s } 138...Qh3+ { +19.41/5 1.01s } 139.Kg6 { -6.20/23 2.34s } 139...Qd3+ { +16.41/5 0.98s } 140.Kf6 { -6.18/19 0.23s } 140...Kc5 { +27.49/5 1.00s } 141.Re6 { -6.21/19 0.66s } 141...Qf3+ { +22.07/5 0.99s } 142.Ke7 { -5.51/19 1.04s } 142...Qf4 { +28.41/5 0.99s } 143.Ra6 { -5.51/19 1.17s } 143...Qc7+ { +18.96/5 0.98s } 144.Kf6 { -5.43/19 0.25s } 144...Kb5 { +15.72/5 1.05s } 145.Re6 { -5.08/15 0.51s } 145...Kc4 { +8.99/5 0.94s } 146.Re5 { -4.69/16 0.57s } 146...Kd4 { +5.96/5 0.92s } 147.Rh5 { -3.47/16 1.30s } 147...Ke4 { +3.55/6 0.98s } 148.Rh1 { -2.05/16 3.56s } 148...Qf4+ { +2.18/6 1.04s } 149.Ke6 { -0.41/14 0.52s } 149...Qg3 { +0.88/6 0.96s } 150.Rb1 { 0.00/15 0.60s } 150...Qc3 { +0.40/6 1.07s } 151.Rh1 { 0.00/14 0.19s } 151...Qg3 { +0.15/6 1.03s } 152.Kd7 { 0.00/22 0.29s } 152...Kd5 { +0.07/5 0.95s } 153.Rd1+ { 0.00/30 0.34s } 153...Kc5 { +0.03/5 1.07s } 154.Rc1+ { 0.00/31 1.43s } 154...Kd5 { 0.00/4 1.02s } 155.Rd1+ { 0.00/102 0.41s } 155...Ke5 { 0.00/2 0.98s } 156.Ra1 { 0.00/245 0.006s, Draw by fifty moves rule } 1/2-1/2[/pgn]

[Ferdy]

Sorry ,
the cuda drivers are older only, the lc0 version is the same:
cublasLt64_11.dll 214.235.648 16.09.2020 05:17 -a-- in the CUDA package
or
cublas64_100.dll 67.318.784 26.08.2018 06:33 -a-- in the CUDNN package

clear now?

We are clear