Files
rustytorch/benchmark_reports/thor-rust-py-pinn-benchmark-12-10-2025/rust_gpu_output.txt
T
2026-03-04 00:08:42 +00:00

587 lines
27 KiB
Plaintext

calculate_k time: [20.444 ns 20.447 ns 20.450 ns]
change: [-0.2596% -0.1009% -0.0106%] (p = 0.13 > 0.05)
No change in performance detected.
Found 3 outliers among 100 measurements (3.00%)
2 (2.00%) high mild
1 (1.00%) high severe
data_generation/synthesize_displacement/200
time: [2.4531 µs 2.4535 µs 2.4539 µs]
thrpt: [81.501 Melem/s 81.515 Melem/s 81.528 Melem/s]
change:
time: [-0.2017% -0.1254% -0.0314%] (p = 0.00 < 0.05)
thrpt: [+0.0314% +0.1256% +0.2021%]
Change within noise threshold.
Found 2 outliers among 100 measurements (2.00%)
1 (1.00%) high mild
1 (1.00%) high severe
data_generation/synthesize_displacement/1000
time: [11.830 µs 11.832 µs 11.833 µs]
thrpt: [84.511 Melem/s 84.520 Melem/s 84.527 Melem/s]
change:
time: [+0.0946% +0.1050% +0.1157%] (p = 0.00 < 0.05)
thrpt: [-0.1156% -0.1048% -0.0945%]
Change within noise threshold.
Found 2 outliers among 100 measurements (2.00%)
1 (1.00%) low mild
1 (1.00%) high severe
data_generation/synthesize_displacement/10000
time: [114.99 µs 115.00 µs 115.01 µs]
thrpt: [86.950 Melem/s 86.958 Melem/s 86.964 Melem/s]
change:
time: [-0.0402% -0.0251% -0.0121%] (p = 0.00 < 0.05)
thrpt: [+0.0121% +0.0251% +0.0402%]
Change within noise threshold.
Found 11 outliers among 100 measurements (11.00%)
2 (2.00%) low mild
6 (6.00%) high mild
3 (3.00%) high severe
data_generation/synthesize_displacement/100000
time: [1.1595 ms 1.1597 ms 1.1599 ms]
thrpt: [86.213 Melem/s 86.231 Melem/s 86.242 Melem/s]
change:
time: [-0.0617% -0.0254% +0.0049%] (p = 0.13 > 0.05)
thrpt: [-0.0049% +0.0254% +0.0617%]
No change in performance detected.
Found 7 outliers among 100 measurements (7.00%)
5 (5.00%) high mild
2 (2.00%) high severe
forward_pass/lffn_mlp/200
time: [1.3635 ms 1.3642 ms 1.3648 ms]
thrpt: [146.54 Kelem/s 146.61 Kelem/s 146.68 Kelem/s]
change:
time: [+35.709% +35.830% +35.952%] (p = 0.00 < 0.05)
thrpt: [-26.444% -26.379% -26.313%]
Performance has regressed.
Found 2 outliers among 50 measurements (4.00%)
1 (2.00%) high mild
1 (2.00%) high severe
forward_pass/lffn_mlp/1000
time: [5.0988 ms 5.1021 ms 5.1056 ms]
thrpt: [195.86 Kelem/s 196.00 Kelem/s 196.13 Kelem/s]
change:
time: [-36.382% -36.310% -36.237%] (p = 0.00 < 0.05)
thrpt: [+56.832% +57.011% +57.189%]
Performance has improved.
Found 7 outliers among 50 measurements (14.00%)
1 (2.00%) low severe
6 (12.00%) low mild
forward_pass/lffn_mlp/10000
time: [48.058 ms 48.165 ms 48.316 ms]
thrpt: [206.97 Kelem/s 207.62 Kelem/s 208.08 Kelem/s]
change:
time: [-42.362% -42.232% -42.054%] (p = 0.00 < 0.05)
thrpt: [+72.574% +73.108% +73.497%]
Performance has improved.
Found 5 outliers among 50 measurements (10.00%)
2 (4.00%) high mild
3 (6.00%) high severe
forward_pass_optimized/lffn_mlp_workspace/200
time: [45.318 µs 45.332 µs 45.345 µs]
thrpt: [4.4106 Melem/s 4.4119 Melem/s 4.4132 Melem/s]
change:
time: [-72.903% -72.816% -72.726%] (p = 0.00 < 0.05)
thrpt: [+266.65% +267.86% +269.05%]
Performance has improved.
Found 1 outliers among 50 measurements (2.00%)
1 (2.00%) high mild
forward_pass_optimized/lffn_mlp_workspace/1000
time: [46.085 µs 46.103 µs 46.121 µs]
thrpt: [21.682 Melem/s 21.690 Melem/s 21.699 Melem/s]
change:
time: [-94.691% -94.668% -94.652%] (p = 0.00 < 0.05)
thrpt: [+1769.7% +1775.5% +1783.7%]
Performance has improved.
Found 7 outliers among 50 measurements (14.00%)
2 (4.00%) low severe
2 (4.00%) low mild
2 (4.00%) high mild
1 (2.00%) high severe
forward_pass_optimized/lffn_mlp_workspace/10000
time: [68.524 µs 68.546 µs 68.566 µs]
thrpt: [145.84 Melem/s 145.89 Melem/s 145.93 Melem/s]
change:
time: [-99.508% -99.492% -99.475%] (p = 0.00 < 0.05)
thrpt: [+18942% +19578% +20226%]
Performance has improved.
Found 1 outliers among 50 measurements (2.00%)
1 (2.00%) low severe
forward_pass_optimized/lffn_mlp_workspace/100000
time: [306.75 µs 306.80 µs 306.84 µs]
thrpt: [325.91 Melem/s 325.95 Melem/s 325.99 Melem/s]
change:
time: [-99.797% -99.794% -99.792%] (p = 0.00 < 0.05)
thrpt: [+48048% +48538% +49048%]
Performance has improved.
Found 5 outliers among 50 measurements (10.00%)
2 (4.00%) low severe
3 (6.00%) low mild
forward_pass_cuda_graph/lffn_mlp_cuda_graph/200
time: [46.444 µs 46.454 µs 46.463 µs]
thrpt: [4.3045 Melem/s 4.3054 Melem/s 4.3063 Melem/s]
Found 3 outliers among 50 measurements (6.00%)
2 (4.00%) low severe
1 (2.00%) high mild
forward_pass_cuda_graph/lffn_mlp_cuda_graph/1000
time: [47.489 µs 47.504 µs 47.518 µs]
thrpt: [21.045 Melem/s 21.051 Melem/s 21.057 Melem/s]
Found 2 outliers among 50 measurements (4.00%)
1 (2.00%) low severe
1 (2.00%) high severe
forward_pass_cuda_graph/lffn_mlp_cuda_graph/10000
time: [69.181 µs 69.202 µs 69.223 µs]
thrpt: [144.46 Melem/s 144.50 Melem/s 144.55 Melem/s]
Found 3 outliers among 50 measurements (6.00%)
1 (2.00%) low severe
2 (4.00%) high mild
forward_pass_matmuls_graph/hybrid_graph/200
time: [41.742 µs 41.752 µs 41.762 µs]
thrpt: [4.7891 Melem/s 4.7902 Melem/s 4.7913 Melem/s]
Found 9 outliers among 100 measurements (9.00%)
3 (3.00%) low severe
2 (2.00%) low mild
4 (4.00%) high mild
forward_pass_matmuls_graph/hybrid_graph/1000
time: [41.707 µs 41.713 µs 41.719 µs]
thrpt: [23.970 Melem/s 23.974 Melem/s 23.977 Melem/s]
Found 4 outliers among 100 measurements (4.00%)
1 (1.00%) low mild
3 (3.00%) high mild
forward_pass_matmuls_graph/hybrid_graph/10000
time: [67.193 µs 67.214 µs 67.233 µs]
thrpt: [148.74 Melem/s 148.78 Melem/s 148.82 Melem/s]
Found 8 outliers among 100 measurements (8.00%)
1 (1.00%) low severe
4 (4.00%) low mild
1 (1.00%) high mild
2 (2.00%) high severe
forward_pass_uber_kernel/uber_value_only/200
time: [332.54 µs 332.59 µs 332.65 µs]
thrpt: [601.24 Kelem/s 601.34 Kelem/s 601.43 Kelem/s]
Found 1 outliers among 100 measurements (1.00%)
1 (1.00%) low severe
forward_pass_uber_kernel/uber_with_gradients/200
time: [590.93 µs 591.06 µs 591.20 µs]
thrpt: [338.29 Kelem/s 338.37 Kelem/s 338.45 Kelem/s]
Found 5 outliers among 100 measurements (5.00%)
2 (2.00%) low severe
1 (1.00%) low mild
2 (2.00%) high mild
forward_pass_uber_kernel/uber_value_only/1000
time: [345.45 µs 345.46 µs 345.48 µs]
thrpt: [2.8946 Melem/s 2.8947 Melem/s 2.8948 Melem/s]
Found 9 outliers among 100 measurements (9.00%)
3 (3.00%) low severe
3 (3.00%) low mild
3 (3.00%) high severe
forward_pass_uber_kernel/uber_with_gradients/1000
time: [607.77 µs 607.87 µs 607.98 µs]
thrpt: [1.6448 Melem/s 1.6451 Melem/s 1.6454 Melem/s]
Found 5 outliers among 100 measurements (5.00%)
2 (2.00%) low severe
3 (3.00%) high mild
forward_pass_uber_kernel/uber_value_only/10000
time: [346.90 µs 346.91 µs 346.93 µs]
thrpt: [28.825 Melem/s 28.826 Melem/s 28.827 Melem/s]
Found 6 outliers among 100 measurements (6.00%)
2 (2.00%) low severe
1 (1.00%) low mild
1 (1.00%) high mild
2 (2.00%) high severe
forward_pass_uber_kernel/uber_with_gradients/10000
time: [616.27 µs 616.34 µs 616.40 µs]
thrpt: [16.223 Melem/s 16.225 Melem/s 16.227 Melem/s]
Found 4 outliers among 100 measurements (4.00%)
1 (1.00%) low severe
1 (1.00%) low mild
1 (1.00%) high mild
1 (1.00%) high severe
forward_pass_uber_kernel_zero_alloc/uber_zero_alloc/200
time: [329.89 µs 329.89 µs 329.89 µs]
thrpt: [606.26 Kelem/s 606.26 Kelem/s 606.27 Kelem/s]
Found 37 outliers among 200 measurements (18.50%)
11 (5.50%) low severe
6 (3.00%) low mild
10 (5.00%) high mild
10 (5.00%) high severe
forward_pass_uber_kernel_zero_alloc/uber_zero_alloc/1000
time: [342.35 µs 342.36 µs 342.38 µs]
thrpt: [2.9208 Melem/s 2.9209 Melem/s 2.9210 Melem/s]
Found 16 outliers among 200 measurements (8.00%)
5 (2.50%) low severe
5 (2.50%) low mild
1 (0.50%) high mild
5 (2.50%) high severe
forward_pass_uber_kernel_zero_alloc/uber_zero_alloc/10000
time: [343.81 µs 343.81 µs 343.82 µs]
thrpt: [29.085 Melem/s 29.086 Melem/s 29.086 Melem/s]
Found 19 outliers among 200 measurements (9.50%)
3 (1.50%) low severe
4 (2.00%) low mild
9 (4.50%) high mild
3 (1.50%) high severe
forward_pass_fused_lite/fused_lite/200
time: [261.04 µs 261.09 µs 261.14 µs]
thrpt: [765.86 Kelem/s 766.02 Kelem/s 766.16 Kelem/s]
Found 1 outliers among 200 measurements (0.50%)
1 (0.50%) high mild
forward_pass_fused_lite/fused_lite/1000
time: [266.63 µs 266.66 µs 266.69 µs]
thrpt: [3.7497 Melem/s 3.7501 Melem/s 3.7505 Melem/s]
Found 16 outliers among 200 measurements (8.00%)
4 (2.00%) low severe
5 (2.50%) low mild
2 (1.00%) high mild
5 (2.50%) high severe
forward_pass_fused_lite/fused_lite/10000
time: [272.84 µs 272.85 µs 272.87 µs]
thrpt: [36.648 Melem/s 36.650 Melem/s 36.652 Melem/s]
Found 12 outliers among 200 measurements (6.00%)
3 (1.50%) low severe
5 (2.50%) low mild
4 (2.00%) high mild
matmul_tensor_cores/fp32_sgemm/200x128x64
time: [19.679 µs 19.683 µs 19.687 µs]
thrpt: [166.44 Gelem/s 166.48 Gelem/s 166.51 Gelem/s]
Found 6 outliers among 100 measurements (6.00%)
2 (2.00%) low mild
3 (3.00%) high mild
1 (1.00%) high severe
matmul_tensor_cores/fp16_tensor_cores/200x128x64
time: [17.718 µs 17.721 µs 17.724 µs]
thrpt: [184.88 Gelem/s 184.91 Gelem/s 184.94 Gelem/s]
Found 3 outliers among 100 measurements (3.00%)
1 (1.00%) low mild
2 (2.00%) high mild
matmul_tensor_cores/fp32_sgemm/200x64x64
time: [19.754 µs 19.758 µs 19.762 µs]
thrpt: [82.907 Gelem/s 82.923 Gelem/s 82.939 Gelem/s]
Found 9 outliers among 100 measurements (9.00%)
2 (2.00%) low mild
5 (5.00%) high mild
2 (2.00%) high severe
matmul_tensor_cores/fp16_tensor_cores/200x64x64
time: [17.811 µs 17.814 µs 17.816 µs]
thrpt: [91.960 Gelem/s 91.975 Gelem/s 91.989 Gelem/s]
Found 2 outliers among 100 measurements (2.00%)
2 (2.00%) high mild
matmul_tensor_cores/fp32_sgemm/1000x128x64
time: [21.346 µs 21.351 µs 21.357 µs]
thrpt: [767.16 Gelem/s 767.36 Gelem/s 767.56 Gelem/s]
Found 13 outliers among 100 measurements (13.00%)
5 (5.00%) low severe
4 (4.00%) low mild
2 (2.00%) high mild
2 (2.00%) high severe
matmul_tensor_cores/fp16_tensor_cores/1000x128x64
time: [17.878 µs 17.886 µs 17.894 µs]
thrpt: [915.63 Gelem/s 916.01 Gelem/s 916.42 Gelem/s]
Found 1 outliers among 100 measurements (1.00%)
1 (1.00%) high severe
matmul_tensor_cores/fp32_sgemm/1000x64x64
time: [18.837 µs 18.842 µs 18.848 µs]
thrpt: [434.64 Gelem/s 434.76 Gelem/s 434.89 Gelem/s]
Found 9 outliers among 100 measurements (9.00%)
3 (3.00%) low severe
4 (4.00%) low mild
1 (1.00%) high mild
1 (1.00%) high severe
matmul_tensor_cores/fp16_tensor_cores/1000x64x64
time: [17.858 µs 17.861 µs 17.863 µs]
thrpt: [458.59 Gelem/s 458.66 Gelem/s 458.73 Gelem/s]
Found 1 outliers among 100 measurements (1.00%)
1 (1.00%) high severe
matmul_tensor_cores/fp32_sgemm/10000x128x64
time: [28.129 µs 28.138 µs 28.146 µs]
thrpt: [5821.2 Gelem/s 5822.8 Gelem/s 5824.5 Gelem/s]
Found 5 outliers among 100 measurements (5.00%)
1 (1.00%) low severe
2 (2.00%) low mild
2 (2.00%) high mild
matmul_tensor_cores/fp16_tensor_cores/10000x128x64
time: [20.381 µs 20.389 µs 20.397 µs]
thrpt: [8032.6 Gelem/s 8035.7 Gelem/s 8038.9 Gelem/s]
Found 6 outliers among 100 measurements (6.00%)
4 (4.00%) low mild
1 (1.00%) high mild
1 (1.00%) high severe
pde_residual_analytical/helmholtz_cpu/200
time: [5.2211 µs 5.2224 µs 5.2238 µs]
thrpt: [38.287 Melem/s 38.297 Melem/s 38.306 Melem/s]
change:
time: [+0.0170% +0.0454% +0.0769%] (p = 0.00 < 0.05)
thrpt: [-0.0768% -0.0454% -0.0170%]
Change within noise threshold.
pde_residual_analytical/helmholtz_cpu/1000
time: [25.163 µs 25.164 µs 25.164 µs]
thrpt: [39.739 Melem/s 39.740 Melem/s 39.741 Melem/s]
change:
time: [-0.0626% -0.0350% -0.0158%] (p = 0.00 < 0.05)
thrpt: [+0.0158% +0.0350% +0.0626%]
Change within noise threshold.
Found 3 outliers among 50 measurements (6.00%)
1 (2.00%) low mild
1 (2.00%) high mild
1 (2.00%) high severe
pde_residual_analytical/helmholtz_cpu/10000
time: [245.36 µs 245.37 µs 245.38 µs]
thrpt: [40.753 Melem/s 40.755 Melem/s 40.757 Melem/s]
change:
time: [+0.3082% +0.3181% +0.3303%] (p = 0.00 < 0.05)
thrpt: [-0.3292% -0.3171% -0.3073%]
Change within noise threshold.
Found 5 outliers among 50 measurements (10.00%)
1 (2.00%) low severe
3 (6.00%) high mild
1 (2.00%) high severe
pde_residual_tensor/helmholtz_tensor/200
time: [177.85 µs 177.90 µs 177.96 µs]
thrpt: [1.1238 Melem/s 1.1242 Melem/s 1.1245 Melem/s]
change:
time: [+1877.5% +1878.6% +1879.7%] (p = 0.00 < 0.05)
thrpt: [-94.949% -94.946% -94.943%]
Performance has regressed.
Found 6 outliers among 50 measurements (12.00%)
1 (2.00%) low severe
3 (6.00%) low mild
1 (2.00%) high mild
1 (2.00%) high severe
pde_residual_tensor/helmholtz_tensor/1000
time: [240.61 µs 240.75 µs 240.89 µs]
thrpt: [4.1513 Melem/s 4.1536 Melem/s 4.1560 Melem/s]
change:
time: [+639.97% +641.03% +641.73%] (p = 0.00 < 0.05)
thrpt: [-86.518% -86.505% -86.486%]
Performance has regressed.
Found 1 outliers among 50 measurements (2.00%)
1 (2.00%) low severe
pde_residual_tensor/helmholtz_tensor/10000
time: [598.41 µs 598.60 µs 598.79 µs]
thrpt: [16.700 Melem/s 16.706 Melem/s 16.711 Melem/s]
change:
time: [+108.95% +109.04% +109.15%] (p = 0.00 < 0.05)
thrpt: [-52.186% -52.161% -52.141%]
Performance has regressed.
Found 1 outliers among 50 measurements (2.00%)
1 (2.00%) high severe
mse_loss/mse_computation/200
time: [50.598 µs 50.617 µs 50.634 µs]
thrpt: [7.8999 Melem/s 7.9025 Melem/s 7.9054 Melem/s]
change:
time: [+3960.9% +3963.6% +3966.0%] (p = 0.00 < 0.05)
thrpt: [-97.541% -97.539% -97.537%]
Performance has regressed.
Found 3 outliers among 50 measurements (6.00%)
1 (2.00%) low severe
2 (4.00%) low mild
mse_loss/mse_computation/1000
time: [61.058 µs 61.107 µs 61.155 µs]
thrpt: [32.704 Melem/s 32.730 Melem/s 32.756 Melem/s]
change:
time: [+2256.2% +2258.5% +2261.0%] (p = 0.00 < 0.05)
thrpt: [-95.764% -95.760% -95.756%]
Performance has regressed.
mse_loss/mse_computation/10000
time: [118.77 µs 118.81 µs 118.85 µs]
thrpt: [168.28 Melem/s 168.33 Melem/s 168.39 Melem/s]
change:
time: [+489.57% +490.01% +490.52%] (p = 0.00 < 0.05)
thrpt: [-83.066% -83.051% -83.038%]
Performance has regressed.
Found 2 outliers among 50 measurements (4.00%)
1 (2.00%) high mild
1 (2.00%) high severe
training_step/single_step/200
time: [1.6020 ms 1.6025 ms 1.6031 ms]
thrpt: [124.76 Kelem/s 124.80 Kelem/s 124.85 Kelem/s]
change:
time: [+53.952% +54.130% +54.302%] (p = 0.00 < 0.05)
thrpt: [-35.192% -35.120% -35.045%]
Performance has regressed.
Found 3 outliers among 30 measurements (10.00%)
3 (10.00%) low mild
training_step/single_step/1000
time: [5.3790 ms 5.3831 ms 5.3875 ms]
thrpt: [185.62 Kelem/s 185.77 Kelem/s 185.91 Kelem/s]
change:
time: [-33.396% -33.149% -32.955%] (p = 0.00 < 0.05)
thrpt: [+49.153% +49.586% +50.141%]
Performance has improved.
Found 2 outliers among 30 measurements (6.67%)
1 (3.33%) high mild
1 (3.33%) high severe
training_step_optimized/single_step_workspace/200
time: [288.90 µs 289.06 µs 289.21 µs]
thrpt: [691.53 Kelem/s 691.90 Kelem/s 692.28 Kelem/s]
change:
time: [+68.047% +68.707% +69.331%] (p = 0.00 < 0.05)
thrpt: [-40.944% -40.726% -40.493%]
Performance has regressed.
training_step_optimized/single_step_workspace/1000
time: [352.56 µs 352.74 µs 352.94 µs]
thrpt: [2.8334 Melem/s 2.8349 Melem/s 2.8364 Melem/s]
change:
time: [-61.113% -60.815% -60.622%] (p = 0.00 < 0.05)
thrpt: [+153.95% +155.20% +157.15%]
Performance has improved.
Found 2 outliers among 30 measurements (6.67%)
1 (3.33%) low severe
1 (3.33%) low mild
training_step_cached/single_step_cached/200
time: [1.5824 ms 1.5829 ms 1.5835 ms]
thrpt: [126.31 Kelem/s 126.35 Kelem/s 126.39 Kelem/s]
change:
time: [+55.274% +55.412% +55.544%] (p = 0.00 < 0.05)
thrpt: [-35.709% -35.655% -35.598%]
Performance has regressed.
Found 3 outliers among 30 measurements (10.00%)
1 (3.33%) high mild
2 (6.67%) high severe
training_step_cached/single_step_cached/1000
time: [5.3466 ms 5.3529 ms 5.3596 ms]
thrpt: [186.58 Kelem/s 186.81 Kelem/s 187.03 Kelem/s]
change:
time: [-33.297% -33.201% -33.114%] (p = 0.00 < 0.05)
thrpt: [+49.508% +49.703% +49.918%]
Performance has improved.
Found 1 outliers among 30 measurements (3.33%)
1 (3.33%) high mild
training_step_fully_optimized/single_step_fully_opt/200
time: [268.67 µs 268.83 µs 268.99 µs]
thrpt: [743.52 Kelem/s 743.97 Kelem/s 744.41 Kelem/s]
change:
time: [+62.589% +63.069% +63.656%] (p = 0.00 < 0.05)
thrpt: [-38.896% -38.676% -38.495%]
Performance has regressed.
Found 1 outliers among 30 measurements (3.33%)
1 (3.33%) high mild
training_step_fully_optimized/single_step_fully_opt/1000
time: [312.08 µs 312.29 µs 312.48 µs]
thrpt: [3.2002 Melem/s 3.2022 Melem/s 3.2043 Melem/s]
change:
time: [-63.992% -63.932% -63.867%] (p = 0.00 < 0.05)
thrpt: [+176.75% +177.25% +177.72%]
Performance has improved.
training_step_deferred_loss/no_loss_step/200
time: [45.941 µs 45.949 µs 45.957 µs]
thrpt: [4.3519 Melem/s 4.3526 Melem/s 4.3534 Melem/s]
training_step_deferred_loss/with_loss_step/200
time: [269.24 µs 269.39 µs 269.56 µs]
thrpt: [741.95 Kelem/s 742.43 Kelem/s 742.82 Kelem/s]
Found 1 outliers among 30 measurements (3.33%)
1 (3.33%) high mild
training_step_deferred_loss/no_loss_step/1000
time: [47.423 µs 47.440 µs 47.457 µs]
thrpt: [21.072 Melem/s 21.079 Melem/s 21.087 Melem/s]
Found 2 outliers among 30 measurements (6.67%)
1 (3.33%) low severe
1 (3.33%) high mild
training_step_deferred_loss/with_loss_step/1000
time: [312.35 µs 312.53 µs 312.73 µs]
thrpt: [3.1976 Melem/s 3.1997 Melem/s 3.2015 Melem/s]
Found 1 outliers among 30 measurements (3.33%)
1 (3.33%) high severe
training_step_zero_sync/single_step_zero_sync/200
time: [346.53 µs 346.74 µs 346.91 µs]
thrpt: [576.52 Kelem/s 576.80 Kelem/s 577.14 Kelem/s]
Found 2 outliers among 30 measurements (6.67%)
1 (3.33%) low mild
1 (3.33%) high severe
training_step_zero_sync/single_step_zero_sync/1000
time: [389.64 µs 390.01 µs 390.38 µs]
thrpt: [2.5616 Melem/s 2.5640 Melem/s 2.5665 Melem/s]
Found 3 outliers among 30 measurements (10.00%)
1 (3.33%) low mild
2 (6.67%) high mild
training_step_cuda_graph/single_step_cuda_graph/200
time: [269.01 µs 269.26 µs 269.43 µs]
thrpt: [742.30 Kelem/s 742.77 Kelem/s 743.46 Kelem/s]
Found 5 outliers among 30 measurements (16.67%)
4 (13.33%) low mild
1 (3.33%) high severe
training_step_cuda_graph/single_step_cuda_graph/1000
time: [311.87 µs 312.07 µs 312.25 µs]
thrpt: [3.2025 Melem/s 3.2044 Melem/s 3.2065 Melem/s]
Found 1 outliers among 30 measurements (3.33%)
1 (3.33%) high severe
training_100_epochs/train_100/200
time: [161.29 ms 161.36 ms 161.42 ms]
thrpt: [619.48 elem/s 619.72 elem/s 620.00 elem/s]
change:
time: [+54.470% +54.742% +55.007%] (p = 0.00 < 0.05)
thrpt: [-35.487% -35.376% -35.263%]
Performance has regressed.
training_100_epochs_cached/train_100_cached/200
time: [159.46 ms 159.56 ms 159.73 ms]
thrpt: [626.05 elem/s 626.73 elem/s 627.12 elem/s]
change:
time: [+54.743% +55.144% +55.504%] (p = 0.00 < 0.05)
thrpt: [-35.693% -35.544% -35.377%]
Performance has regressed.
Found 1 outliers among 10 measurements (10.00%)
1 (10.00%) high mild
training_100_epochs_fully_optimized/train_100_fully_opt/200
time: [28.016 ms 28.030 ms 28.040 ms]
thrpt: [3.5663 Kelem/s 3.5677 Kelem/s 3.5694 Kelem/s]
change:
time: [+61.278% +63.573% +65.974%] (p = 0.00 < 0.05)
thrpt: [-39.750% -38.865% -37.995%]
Performance has regressed.
Found 1 outliers among 10 measurements (10.00%)
1 (10.00%) low mild
training_100_epochs_deferred_loss/train_100_deferred_loss/200
time: [5.8959 ms 5.8975 ms 5.8997 ms]
thrpt: [16.950 Kelem/s 16.956 Kelem/s 16.961 Kelem/s]
training_100_epochs_zero_sync/train_100_zero_sync/200
time: [35.784 ms 35.806 ms 35.824 ms]
thrpt: [2.7914 Kelem/s 2.7929 Kelem/s 2.7945 Kelem/s]
training_100_epochs_cuda_graph/train_100_cuda_graph/200
time: [28.010 ms 28.020 ms 28.035 ms]
thrpt: [3.5669 Kelem/s 3.5689 Kelem/s 3.5701 Kelem/s]
training_step_analytical/analytical_backprop/200
time: [186.14 µs 186.19 µs 186.25 µs]
thrpt: [1.0738 Melem/s 1.0742 Melem/s 1.0745 Melem/s]
Found 13 outliers among 200 measurements (6.50%)
2 (1.00%) low severe
6 (3.00%) low mild
4 (2.00%) high mild
1 (0.50%) high severe
training_step_analytical/analytical_backprop/1000
time: [300.75 µs 300.83 µs 300.91 µs]
thrpt: [3.3233 Melem/s 3.3242 Melem/s 3.3250 Melem/s]
Found 23 outliers among 200 measurements (11.50%)
4 (2.00%) low severe
8 (4.00%) low mild
8 (4.00%) high mild
3 (1.50%) high severe
training_100_epochs_analytical/train_100_analytical/200
time: [17.807 ms 17.813 ms 17.819 ms]
thrpt: [5.6120 Kelem/s 5.6139 Kelem/s 5.6156 Kelem/s]