From: Georgi Gerganov Date: Thu, 15 Jan 2026 09:53:09 +0000 (+0200) Subject: benches : update X-Git-Tag: upstream/1.8.3~1 X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=21c1765fcb562cbc13e31f80bf9c9a4961d4f04a;p=pkg%2Fggml%2Fsources%2Fwhisper.cpp benches : update --- diff --git a/scripts/bench-all-gg.txt b/scripts/bench-all-gg.txt index cf3d26fb..32a09083 100644 --- a/scripts/bench-all-gg.txt +++ b/scripts/bench-all-gg.txt @@ -111,61 +111,61 @@ make -j && ./scripts/bench-all.sh 1 1 0 | CPU | Config | Model | Th | FA | Enc. | Dec. | Bch5 | PP | Commit | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| M2 ULTRA | METAL | tiny | 1 | 0 | 8.82 | 1.14 | 0.28 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | tiny-q5_0 | 1 | 0 | 9.28 | 1.11 | 0.29 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | tiny-q5_1 | 1 | 0 | 9.28 | 1.11 | 0.29 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | tiny-q8_0 | 1 | 0 | 8.94 | 1.12 | 0.28 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | base | 1 | 0 | 15.84 | 1.60 | 0.43 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | base-q5_0 | 1 | 0 | 17.62 | 1.61 | 0.47 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | base-q5_1 | 1 | 0 | 17.00 | 1.57 | 0.45 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | base-q8_0 | 1 | 0 | 16.19 | 1.56 | 0.43 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | small | 1 | 0 | 47.72 | 3.12 | 0.92 | 0.06 | 2ad7a695 | -| M2 ULTRA | METAL | small-q5_0 | 1 | 0 | 52.59 | 3.13 | 0.94 | 0.06 | 2ad7a695 | -| M2 ULTRA | METAL | small-q5_1 | 1 | 0 | 52.50 | 3.09 | 0.94 | 0.06 | 2ad7a695 | -| M2 ULTRA | METAL | small-q8_0 | 1 | 0 | 48.92 | 2.92 | 0.91 | 0.06 | 2ad7a695 | -| M2 ULTRA | METAL | medium | 1 | 0 | 136.84 | 6.64 | 2.06 | 0.13 | 2ad7a695 | -| M2 ULTRA | METAL | medium-q5_0 | 1 | 0 | 152.83 | 6.32 | 2.13 | 0.14 | 2ad7a695 | -| M2 ULTRA | METAL | medium-q5_1 | 1 | 0 | 153.27 | 6.30 | 2.14 | 0.14 | 2ad7a695 | -| M2 ULTRA | METAL | medium-q8_0 | 1 | 0 | 142.05 | 6.14 | 2.08 | 0.13 | 2ad7a695 | -| M2 ULTRA | METAL | medium-dis | 1 | 0 | 123.80 | 0.91 | 0.25 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2 | 1 | 0 | 238.97 | 9.69 | 3.13 | 0.22 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2-q5_0 | 1 | 0 | 273.72 | 9.31 | 3.17 | 0.25 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2-q5_1 | 1 | 0 | 273.42 | 9.26 | 3.18 | 0.25 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2-q8_0 | 1 | 0 | 247.80 | 9.33 | 3.04 | 0.23 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2-dis | 1 | 0 | 213.83 | 1.00 | 0.28 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | large-v3-turbo | 1 | 0 | 215.47 | 1.54 | 0.47 | 0.03 | 2ad7a695 | -| M2 ULTRA | METAL | large-v3-turbo-q5_0 | 1 | 0 | 246.32 | 1.44 | 0.47 | 0.04 | 2ad7a695 | -| M2 ULTRA | METAL | large-v3-turbo-q8_0 | 1 | 0 | 223.43 | 1.44 | 0.45 | 0.04 | 2ad7a695 | +| M2 ULTRA | METAL | tiny | 1 | 0 | 8.80 | 1.13 | 0.28 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | tiny-q5_0 | 1 | 0 | 9.34 | 1.09 | 0.28 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | tiny-q5_1 | 1 | 0 | 9.29 | 1.09 | 0.29 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | tiny-q8_0 | 1 | 0 | 9.00 | 1.12 | 0.28 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | base | 1 | 0 | 15.92 | 1.60 | 0.43 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | base-q5_0 | 1 | 0 | 17.01 | 1.53 | 0.43 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | base-q5_1 | 1 | 0 | 17.02 | 1.53 | 0.44 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | base-q8_0 | 1 | 0 | 16.25 | 1.55 | 0.43 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | small | 1 | 0 | 47.83 | 3.09 | 0.91 | 0.05 | 47af2fb7 | +| M2 ULTRA | METAL | small-q5_0 | 1 | 0 | 52.85 | 2.98 | 0.94 | 0.06 | 47af2fb7 | +| M2 ULTRA | METAL | small-q5_1 | 1 | 0 | 52.92 | 2.97 | 0.94 | 0.06 | 47af2fb7 | +| M2 ULTRA | METAL | small-q8_0 | 1 | 0 | 49.05 | 2.89 | 0.90 | 0.06 | 47af2fb7 | +| M2 ULTRA | METAL | medium | 1 | 0 | 127.98 | 6.62 | 2.05 | 0.12 | 47af2fb7 | +| M2 ULTRA | METAL | medium-q5_0 | 1 | 0 | 145.42 | 6.09 | 2.12 | 0.14 | 47af2fb7 | +| M2 ULTRA | METAL | medium-q5_1 | 1 | 0 | 145.16 | 6.08 | 2.14 | 0.14 | 47af2fb7 | +| M2 ULTRA | METAL | medium-q8_0 | 1 | 0 | 132.72 | 6.10 | 2.07 | 0.13 | 47af2fb7 | +| M2 ULTRA | METAL | medium-dis | 1 | 0 | 115.09 | 0.91 | 0.25 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2 | 1 | 0 | 243.69 | 9.68 | 3.14 | 0.22 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2-q5_0 | 1 | 0 | 280.38 | 8.95 | 3.18 | 0.25 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2-q5_1 | 1 | 0 | 279.76 | 8.92 | 3.18 | 0.25 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2-q8_0 | 1 | 0 | 254.55 | 9.35 | 3.04 | 0.23 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2-dis | 1 | 0 | 219.23 | 1.01 | 0.28 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | large-v3-turbo | 1 | 0 | 220.57 | 1.55 | 0.46 | 0.03 | 47af2fb7 | +| M2 ULTRA | METAL | large-v3-turbo-q5_0 | 1 | 0 | 253.03 | 1.40 | 0.47 | 0.04 | 47af2fb7 | +| M2 ULTRA | METAL | large-v3-turbo-q8_0 | 1 | 0 | 229.82 | 1.43 | 0.45 | 0.04 | 47af2fb7 | make -j && ./scripts/bench-all.sh 1 1 1 | CPU | Config | Model | Th | FA | Enc. | Dec. | Bch5 | PP | Commit | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| M2 ULTRA | METAL | tiny | 1 | 1 | 6.13 | 0.95 | 0.22 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | tiny-q5_0 | 1 | 1 | 6.56 | 0.91 | 0.22 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | tiny-q5_1 | 1 | 1 | 6.59 | 0.92 | 0.23 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | tiny-q8_0 | 1 | 1 | 6.23 | 0.93 | 0.22 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | base | 1 | 1 | 10.73 | 1.31 | 0.33 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | base-q5_0 | 1 | 1 | 11.89 | 1.25 | 0.34 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | base-q5_1 | 1 | 1 | 11.83 | 1.24 | 0.34 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | base-q8_0 | 1 | 1 | 11.03 | 1.25 | 0.32 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | small | 1 | 1 | 32.05 | 2.42 | 0.65 | 0.04 | 2ad7a695 | -| M2 ULTRA | METAL | small-q5_0 | 1 | 1 | 36.73 | 2.41 | 0.67 | 0.04 | 2ad7a695 | -| M2 ULTRA | METAL | small-q5_1 | 1 | 1 | 36.77 | 2.41 | 0.68 | 0.04 | 2ad7a695 | -| M2 ULTRA | METAL | small-q8_0 | 1 | 1 | 33.33 | 2.28 | 0.65 | 0.04 | 2ad7a695 | -| M2 ULTRA | METAL | medium | 1 | 1 | 88.19 | 5.10 | 1.47 | 0.09 | 2ad7a695 | -| M2 ULTRA | METAL | medium-q5_0 | 1 | 1 | 104.23 | 4.90 | 1.48 | 0.10 | 2ad7a695 | -| M2 ULTRA | METAL | medium-q5_1 | 1 | 1 | 104.19 | 5.02 | 1.51 | 0.10 | 2ad7a695 | -| M2 ULTRA | METAL | medium-q8_0 | 1 | 1 | 92.41 | 4.96 | 1.44 | 0.09 | 2ad7a695 | -| M2 ULTRA | METAL | medium-dis | 1 | 1 | 76.97 | 0.79 | 0.20 | 0.01 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2 | 1 | 1 | 169.61 | 7.48 | 2.14 | 0.17 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2-q5_0 | 1 | 1 | 203.04 | 7.35 | 2.18 | 0.20 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2-q5_1 | 1 | 1 | 202.91 | 7.32 | 2.20 | 0.20 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2-q8_0 | 1 | 1 | 178.30 | 6.86 | 2.12 | 0.18 | 2ad7a695 | -| M2 ULTRA | METAL | large-v2-dis | 1 | 1 | 146.47 | 0.89 | 0.22 | 0.02 | 2ad7a695 | -| M2 ULTRA | METAL | large-v3-turbo | 1 | 1 | 147.86 | 1.30 | 0.34 | 0.03 | 2ad7a695 | -| M2 ULTRA | METAL | large-v3-turbo-q5_0 | 1 | 1 | 177.75 | 1.17 | 0.35 | 0.03 | 2ad7a695 | -| M2 ULTRA | METAL | large-v3-turbo-q8_0 | 1 | 1 | 155.51 | 1.18 | 0.33 | 0.03 | 2ad7a695 | +| M2 ULTRA | METAL | tiny | 1 | 1 | 6.19 | 0.93 | 0.21 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | tiny-q5_0 | 1 | 1 | 6.64 | 0.89 | 0.22 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | tiny-q5_1 | 1 | 1 | 6.65 | 0.91 | 0.23 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | tiny-q8_0 | 1 | 1 | 6.26 | 0.93 | 0.22 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | base | 1 | 1 | 10.89 | 1.31 | 0.32 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | base-q5_0 | 1 | 1 | 12.10 | 1.22 | 0.33 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | base-q5_1 | 1 | 1 | 12.05 | 1.22 | 0.33 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | base-q8_0 | 1 | 1 | 11.24 | 1.24 | 0.32 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | small | 1 | 1 | 32.06 | 2.41 | 0.64 | 0.04 | 47af2fb7 | +| M2 ULTRA | METAL | small-q5_0 | 1 | 1 | 37.20 | 2.32 | 0.67 | 0.04 | 47af2fb7 | +| M2 ULTRA | METAL | small-q5_1 | 1 | 1 | 37.13 | 2.30 | 0.67 | 0.04 | 47af2fb7 | +| M2 ULTRA | METAL | small-q8_0 | 1 | 1 | 33.63 | 2.28 | 0.64 | 0.04 | 47af2fb7 | +| M2 ULTRA | METAL | medium | 1 | 1 | 89.22 | 5.14 | 1.46 | 0.09 | 47af2fb7 | +| M2 ULTRA | METAL | medium-q5_0 | 1 | 1 | 106.82 | 4.83 | 1.49 | 0.11 | 47af2fb7 | +| M2 ULTRA | METAL | medium-q5_1 | 1 | 1 | 106.60 | 4.88 | 1.50 | 0.11 | 47af2fb7 | +| M2 ULTRA | METAL | medium-q8_0 | 1 | 1 | 94.48 | 4.93 | 1.43 | 0.09 | 47af2fb7 | +| M2 ULTRA | METAL | medium-dis | 1 | 1 | 77.85 | 0.80 | 0.20 | 0.01 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2 | 1 | 1 | 170.73 | 7.50 | 2.12 | 0.16 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2-q5_0 | 1 | 1 | 206.46 | 7.05 | 2.17 | 0.20 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2-q5_1 | 1 | 1 | 206.15 | 7.10 | 2.19 | 0.20 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2-q8_0 | 1 | 1 | 180.31 | 6.90 | 2.10 | 0.17 | 47af2fb7 | +| M2 ULTRA | METAL | large-v2-dis | 1 | 1 | 147.44 | 0.90 | 0.22 | 0.02 | 47af2fb7 | +| M2 ULTRA | METAL | large-v3-turbo | 1 | 1 | 148.79 | 1.30 | 0.34 | 0.03 | 47af2fb7 | +| M2 ULTRA | METAL | large-v3-turbo-q5_0 | 1 | 1 | 180.34 | 1.14 | 0.35 | 0.03 | 47af2fb7 | +| M2 ULTRA | METAL | large-v3-turbo-q8_0 | 1 | 1 | 158.04 | 1.18 | 0.33 | 0.03 | 47af2fb7 | ## M4 Max @@ -218,17 +218,17 @@ make -j && ./scripts/bench-all.sh 1 1 0 | CPU | Config | Model | Th | FA | Enc. | Dec. | Bch5 | PP | Commit | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| M4 Max | METAL | tiny | 1 | 0 | 10.51 | 0.86 | 0.23 | 0.01 | 47fcd7da | -| M4 Max | METAL | tiny-q8_0 | 1 | 0 | 10.73 | 0.84 | 0.24 | 0.01 | 47fcd7da | -| M4 Max | METAL | base | 1 | 0 | 19.50 | 1.34 | 0.36 | 0.02 | 47fcd7da | -| M4 Max | METAL | base-q8_0 | 1 | 0 | 20.17 | 1.25 | 0.36 | 0.02 | 47fcd7da | -| M4 Max | METAL | small | 1 | 0 | 61.91 | 2.77 | 0.78 | 0.06 | 47fcd7da | -| M4 Max | METAL | small-q8_0 | 1 | 0 | 64.17 | 2.43 | 0.78 | 0.06 | 47fcd7da | -| M4 Max | METAL | medium | 1 | 0 | 181.50 | 6.44 | 1.85 | 0.15 | 47fcd7da | -| M4 Max | METAL | medium-q8_0 | 1 | 0 | 187.71 | 5.80 | 1.84 | 0.15 | 47fcd7da | -| M4 Max | METAL | large-v2 | 1 | 0 | 335.49 | 10.49 | 3.01 | 0.26 | 47fcd7da | -| M4 Max | METAL | large-v2-q8_0 | 1 | 0 | 349.89 | 8.65 | 2.97 | 0.27 | 47fcd7da | -| M4 Max | METAL | large-v3-turbo | 1 | 0 | 301.34 | 1.83 | 0.49 | 0.04 | 47fcd7da | +| M4 Max | METAL | tiny | 1 | 0 | 10.75 | 0.87 | 0.24 | 0.01 | 47af2fb7 | +| M4 Max | METAL | tiny-q8_0 | 1 | 0 | 11.15 | 0.85 | 0.24 | 0.01 | 47af2fb7 | +| M4 Max | METAL | base | 1 | 0 | 20.12 | 1.34 | 0.36 | 0.02 | 47af2fb7 | +| M4 Max | METAL | base-q8_0 | 1 | 0 | 20.40 | 1.25 | 0.37 | 0.02 | 47af2fb7 | +| M4 Max | METAL | small | 1 | 0 | 63.80 | 2.75 | 0.77 | 0.06 | 47af2fb7 | +| M4 Max | METAL | small-q8_0 | 1 | 0 | 65.46 | 2.43 | 0.77 | 0.06 | 47af2fb7 | +| M4 Max | METAL | medium | 1 | 0 | 184.43 | 6.21 | 1.82 | 0.15 | 47af2fb7 | +| M4 Max | METAL | medium-q8_0 | 1 | 0 | 190.19 | 5.76 | 1.86 | 0.15 | 47af2fb7 | +| M4 Max | METAL | large-v2 | 1 | 0 | 344.05 | 10.64 | 3.07 | 0.26 | 47af2fb7 | +| M4 Max | METAL | large-v2-q8_0 | 1 | 0 | 355.43 | 8.83 | 3.03 | 0.27 | 47af2fb7 | +| M4 Max | METAL | large-v3-turbo | 1 | 0 | 306.64 | 1.82 | 0.49 | 0.04 | 47af2fb7 | make -j && ./scripts/bench-all.sh 1 1 1 @@ -247,6 +247,20 @@ make -j && ./scripts/bench-all.sh 1 1 1 | M4 Max | METAL | large-v2-q8_0 | 1 | 1 | 296.43 | 7.44 | 2.09 | 0.23 | 47fcd7da | | M4 Max | METAL | large-v3-turbo | 1 | 1 | 249.91 | 1.65 | 0.38 | 0.04 | 47fcd7da | +| CPU | Config | Model | Th | FA | Enc. | Dec. | Bch5 | PP | Commit | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| M4 Max | METAL | tiny | 1 | 1 | 8.23 | 0.72 | 0.16 | 0.01 | 47af2fb7 | +| M4 Max | METAL | tiny-q8_0 | 1 | 1 | 8.38 | 0.68 | 0.16 | 0.01 | 47af2fb7 | +| M4 Max | METAL | base | 1 | 1 | 15.66 | 1.16 | 0.26 | 0.02 | 47af2fb7 | +| M4 Max | METAL | base-q8_0 | 1 | 1 | 15.88 | 1.08 | 0.27 | 0.02 | 47af2fb7 | +| M4 Max | METAL | small | 1 | 1 | 50.34 | 2.38 | 0.54 | 0.05 | 47af2fb7 | +| M4 Max | METAL | small-q8_0 | 1 | 1 | 51.90 | 1.98 | 0.54 | 0.05 | 47af2fb7 | +| M4 Max | METAL | medium | 1 | 1 | 149.55 | 5.59 | 1.30 | 0.12 | 47af2fb7 | +| M4 Max | METAL | medium-q8_0 | 1 | 1 | 154.34 | 4.65 | 1.28 | 0.13 | 47af2fb7 | +| M4 Max | METAL | large-v2 | 1 | 1 | 291.28 | 9.16 | 2.14 | 0.22 | 47af2fb7 | +| M4 Max | METAL | large-v2-q8_0 | 1 | 1 | 301.06 | 7.21 | 2.08 | 0.23 | 47af2fb7 | +| M4 Max | METAL | large-v3-turbo | 1 | 1 | 256.23 | 1.61 | 0.38 | 0.04 | 47af2fb7 | + # RTX 5090 @@ -254,35 +268,72 @@ make -j && ./scripts/bench-all.sh 1 1 0 | GPU | Config | Model | Th | FA | Enc. | Dec. | Bch5 | PP | Commit | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| RTX 5090 | CUDA | tiny | 1 | 0 | 2.06 | 0.55 | 0.13 | 0.00 | e4bf87b0 | -| RTX 5090 | CUDA | tiny-q8_0 | 1 | 0 | 2.50 | 0.55 | 0.14 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | base | 1 | 0 | 3.72 | 0.81 | 0.19 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | base-q8_0 | 1 | 0 | 4.35 | 0.79 | 0.20 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | small | 1 | 0 | 11.24 | 1.55 | 0.38 | 0.02 | e4bf87b0 | -| RTX 5090 | CUDA | small-q8_0 | 1 | 0 | 12.69 | 1.69 | 0.40 | 0.02 | e4bf87b0 | -| RTX 5090 | CUDA | medium | 1 | 0 | 31.16 | 3.19 | 0.79 | 0.04 | e4bf87b0 | -| RTX 5090 | CUDA | medium-q8_0 | 1 | 0 | 32.74 | 3.43 | 0.80 | 0.05 | e4bf87b0 | -| RTX 5090 | CUDA | large-v2 | 1 | 0 | 50.09 | 4.55 | 1.14 | 0.05 | e4bf87b0 | -| RTX 5090 | CUDA | large-v2-q8_0 | 1 | 0 | 52.44 | 4.76 | 1.11 | 0.07 | e4bf87b0 | -| RTX 5090 | CUDA | large-v3-turbo | 1 | 0 | 46.78 | 0.70 | 0.17 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | large-v3-turbo-q8_0 | 1 | 0 | 48.57 | 0.70 | 0.16 | 0.01 | e4bf87b0 | +| RTX 5090 | CUDA | tiny | 1 | 0 | 2.12 | 0.51 | 0.13 | 0.00 | 47af2fb7 | +| RTX 5090 | CUDA | tiny-q8_0 | 1 | 0 | 2.50 | 0.52 | 0.14 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | base | 1 | 0 | 3.74 | 0.76 | 0.19 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | base-q8_0 | 1 | 0 | 4.38 | 0.74 | 0.20 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | small | 1 | 0 | 11.25 | 1.46 | 0.39 | 0.02 | 47af2fb7 | +| RTX 5090 | CUDA | small-q8_0 | 1 | 0 | 12.70 | 1.58 | 0.41 | 0.02 | 47af2fb7 | +| RTX 5090 | CUDA | medium | 1 | 0 | 31.16 | 3.07 | 0.80 | 0.04 | 47af2fb7 | +| RTX 5090 | CUDA | medium-q8_0 | 1 | 0 | 32.50 | 3.23 | 0.83 | 0.05 | 47af2fb7 | +| RTX 5090 | CUDA | large-v2 | 1 | 0 | 50.04 | 4.59 | 1.15 | 0.05 | 47af2fb7 | +| RTX 5090 | CUDA | large-v2-q8_0 | 1 | 0 | 52.17 | 4.38 | 1.14 | 0.07 | 47af2fb7 | +| RTX 5090 | CUDA | large-v3-turbo | 1 | 0 | 46.88 | 0.70 | 0.17 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | large-v3-turbo-q8_0 | 1 | 0 | 48.49 | 0.64 | 0.16 | 0.01 | 47af2fb7 | + +make -j && ./scripts/bench-all.sh 1 1 1 + +| GPU | Config | Model | Th | FA | Enc. | Dec. | Bch5 | PP | Commit | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| RTX 5090 | CUDA | tiny | 1 | 1 | 1.42 | 0.44 | 0.11 | 0.00 | 47af2fb7 | +| RTX 5090 | CUDA | tiny-q8_0 | 1 | 1 | 1.83 | 0.45 | 0.12 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | base | 1 | 1 | 2.21 | 0.65 | 0.16 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | base-q8_0 | 1 | 1 | 2.85 | 0.62 | 0.17 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | small | 1 | 1 | 5.11 | 1.23 | 0.32 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | small-q8_0 | 1 | 1 | 6.50 | 1.35 | 0.34 | 0.02 | 47af2fb7 | +| RTX 5090 | CUDA | medium | 1 | 1 | 14.01 | 2.57 | 0.64 | 0.03 | 47af2fb7 | +| RTX 5090 | CUDA | medium-q8_0 | 1 | 1 | 15.34 | 2.72 | 0.67 | 0.04 | 47af2fb7 | +| RTX 5090 | CUDA | large-v2 | 1 | 1 | 21.70 | 3.96 | 0.97 | 0.04 | 47af2fb7 | +| RTX 5090 | CUDA | large-v2-q8_0 | 1 | 1 | 23.57 | 3.70 | 0.94 | 0.05 | 47af2fb7 | +| RTX 5090 | CUDA | large-v3-turbo | 1 | 1 | 18.61 | 0.62 | 0.15 | 0.01 | 47af2fb7 | +| RTX 5090 | CUDA | large-v3-turbo-q8_0 | 1 | 1 | 20.10 | 0.56 | 0.14 | 0.01 | 47af2fb7 | + + +# DGX Spark + +make -j && ./scripts/bench-all.sh 1 1 0 + +| GPU | Config | Model | Th | FA | Enc. | Dec. | Bch5 | PP | Commit | +| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | +| DGX Spk. | CUDA | tiny | 1 | 0 | 9.42 | 0.85 | 0.22 | 0.01 | 47af2fb7 | +| DGX Spk. | CUDA | tiny-q8_0 | 1 | 0 | 9.69 | 0.81 | 0.20 | 0.01 | 47af2fb7 | +| DGX Spk. | CUDA | base | 1 | 0 | 18.81 | 1.36 | 0.33 | 0.02 | 47af2fb7 | +| DGX Spk. | CUDA | base-q8_0 | 1 | 0 | 18.11 | 1.20 | 0.30 | 0.02 | 47af2fb7 | +| DGX Spk. | CUDA | small | 1 | 0 | 59.83 | 3.01 | 0.74 | 0.04 | 47af2fb7 | +| DGX Spk. | CUDA | small-q8_0 | 1 | 0 | 59.12 | 2.66 | 0.67 | 0.05 | 47af2fb7 | +| DGX Spk. | CUDA | medium | 1 | 0 | 163.73 | 7.53 | 1.70 | 0.12 | 47af2fb7 | +| DGX Spk. | CUDA | medium-q8_0 | 1 | 0 | 157.54 | 5.98 | 1.48 | 0.13 | 47af2fb7 | +| DGX Spk. | CUDA | large-v2 | 1 | 0 | 279.83 | 12.26 | 2.77 | 0.21 | 47af2fb7 | +| DGX Spk. | CUDA | large-v2-q8_0 | 1 | 0 | 273.05 | 9.31 | 2.33 | 0.22 | 47af2fb7 | +| DGX Spk. | CUDA | large-v3-turbo | 1 | 0 | 271.11 | 2.06 | 0.47 | 0.03 | 47af2fb7 | +| DGX Spk. | CUDA | large-v3-turbo-q8_0 | 1 | 0 | 262.69 | 1.49 | 0.36 | 0.03 | 47af2fb7 | make -j && ./scripts/bench-all.sh 1 1 1 | GPU | Config | Model | Th | FA | Enc. | Dec. | Bch5 | PP | Commit | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | -| RTX 5090 | CUDA | tiny | 1 | 1 | 1.39 | 0.47 | 0.11 | 0.00 | e4bf87b0 | -| RTX 5090 | CUDA | tiny-q8_0 | 1 | 1 | 1.83 | 0.48 | 0.12 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | base | 1 | 1 | 2.17 | 0.70 | 0.16 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | base-q8_0 | 1 | 1 | 2.78 | 0.68 | 0.17 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | small | 1 | 1 | 5.02 | 1.33 | 0.32 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | small-q8_0 | 1 | 1 | 6.39 | 1.46 | 0.34 | 0.02 | e4bf87b0 | -| RTX 5090 | CUDA | medium | 1 | 1 | 13.89 | 2.68 | 0.64 | 0.03 | e4bf87b0 | -| RTX 5090 | CUDA | medium-q8_0 | 1 | 1 | 15.40 | 2.92 | 0.67 | 0.04 | e4bf87b0 | -| RTX 5090 | CUDA | large-v2 | 1 | 1 | 21.24 | 3.88 | 0.96 | 0.04 | e4bf87b0 | -| RTX 5090 | CUDA | large-v2-q8_0 | 1 | 1 | 23.54 | 4.01 | 0.93 | 0.05 | e4bf87b0 | -| RTX 5090 | CUDA | large-v3-turbo | 1 | 1 | 18.18 | 0.62 | 0.15 | 0.01 | e4bf87b0 | -| RTX 5090 | CUDA | large-v3-turbo-q8_0 | 1 | 1 | 19.89 | 0.61 | 0.14 | 0.01 | e4bf87b0 | +| DGX Spk. | CUDA | tiny | 1 | 1 | 2.89 | 0.76 | 0.19 | 0.01 | 47af2fb7 | +| DGX Spk. | CUDA | tiny-q8_0 | 1 | 1 | 3.06 | 0.72 | 0.17 | 0.01 | 47af2fb7 | +| DGX Spk. | CUDA | base | 1 | 1 | 5.37 | 1.23 | 0.29 | 0.01 | 47af2fb7 | +| DGX Spk. | CUDA | base-q8_0 | 1 | 1 | 4.70 | 1.07 | 0.26 | 0.01 | 47af2fb7 | +| DGX Spk. | CUDA | small | 1 | 1 | 17.70 | 2.73 | 0.66 | 0.02 | 47af2fb7 | +| DGX Spk. | CUDA | small-q8_0 | 1 | 1 | 16.77 | 2.38 | 0.58 | 0.03 | 47af2fb7 | +| DGX Spk. | CUDA | medium | 1 | 1 | 56.22 | 6.98 | 1.53 | 0.06 | 47af2fb7 | +| DGX Spk. | CUDA | medium-q8_0 | 1 | 1 | 46.39 | 5.46 | 1.28 | 0.07 | 47af2fb7 | +| DGX Spk. | CUDA | large-v2 | 1 | 1 | 100.33 | 11.59 | 2.53 | 0.09 | 47af2fb7 | +| DGX Spk. | CUDA | large-v2-q8_0 | 1 | 1 | 97.28 | 8.60 | 2.10 | 0.10 | 47af2fb7 | +| DGX Spk. | CUDA | large-v3-turbo | 1 | 1 | 92.59 | 2.00 | 0.44 | 0.02 | 47af2fb7 | +| DGX Spk. | CUDA | large-v3-turbo-q8_0 | 1 | 1 | 85.96 | 1.40 | 0.33 | 0.02 | 47af2fb7 | # V100