{ "generated_at": "2026-09-10T02:07:47.398958+00:00", "status": "passed", "jobs": [ { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_dit_rank0_cd8cbc742f5c3f37_v1", "elapsed_seconds": 11.601743698120117, "allocated": "6", "input_sha256": "824a77bcaead8af0dacf908827027966a81c2d964d919eb4200bac0771b4a369", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 4680, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 25.312, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 38.789238, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 20.414169, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 56.807163, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.556988, "unit": "%" }, "dram__bytes_read.sum": { "value": 2.86592, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 21.970688, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 981.21871, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 222.404, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "bdad908cfdd17ca569ae9a401f17d14038fcd825ec086f9a16fb0c9fd8d85ac7", "metrics.csv": "48444dbcd0b825c247c4b7094c7b2e678413c213928495d57a944b3cf18dbc82", "operator.ncu-rep": "0a3f7679447eede906abd5f43050b8d1724741a4a8b312e9879e5815b29b5f56", "run.log": "7767026ccab67208b53c1ba2972c2591d62790fae2ac2849ec62f2c3631a0f4e", "identity.json": "bfe7aff0adaf5c4977a742f0c2f29aea19940230b6578ef11d5303eb0dde0798", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "90c7c7a2e12163b078d9cabc7c0c9a158b095d6063db83c572d3c9289e1dbc47", "guard_own_pid_history.json": "526a813f7ee30828fa40c9c126b3c1010f6ca8c2d252170a96dc029923d1be0f" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_dit_rank0_9296b219534b05e0_v1", "elapsed_seconds": 11.829026460647583, "allocated": "6", "input_sha256": "787b43c2c5ab957d3bd3c45047d646e6e407e506ea71b3249408d5d20bf00b0c", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 4680, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 95.52, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 83.74599, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 17.378666, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 54.057958, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.734277, "unit": "%" }, "dram__bytes_read.sum": { "value": 46.2784, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 33.586688, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 836.108543, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 222.404, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "369c62e668a96131ff716e2d7527937010ddb07ace3c1a1113f9beb7aca7da4c", "metrics.csv": "14e3ba7326a14e47cdd8e4fb9bea081d2c716492736a50e51c69198ce0ac9403", "operator.ncu-rep": "bdf1c2ff221f81014468b98b66384f2d477fcf5bcc1043c9498aa1e093816da4", "run.log": "cebcd9658c3c1b7f5ff157aea95db842dd81f17431d27c6beaf6622249dfbbf8", "identity.json": "a3220f175eac4a4a7092bab84b6123b3d9e4eace972e88165934a3e1434647f9", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "a61118ada8ad9040a34f6d4de53d267a7bd8b9e738d0c5e6a3459caded567ed4", "guard_own_pid_history.json": "abfd1629eaf5a0720299946477d8b1c49416882cf2e23367afac19bc360fc142" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_dit_v1", "elapsed_seconds": 14.154895782470703, "allocated": "0", "input_sha256": "4a5b492ee606c1a2c9fe4844a0c3125609fa599f4c027f56b4329c89d2879b7e", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 4680, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 292.448, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 94.027733, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 25.371858, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 52.627935, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.649706, "unit": "%" }, "dram__bytes_read.sum": { "value": 314.121984, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 42.865664, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.220688, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 222.404, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "369c62e668a96131ff716e2d7527937010ddb07ace3c1a1113f9beb7aca7da4c", "metrics.csv": "e3b23b6928422a2d427cc67fa15c47e2fc6a548d66f2a4bf7b6d2626482cecaa", "operator.ncu-rep": "11e4d03e257ff0eca8872072b023ba7829ac2b6b25569f0a020e641022958844", "run.log": "86ac88cc3a8cc99c3e583d48e0b6b129aeea977cf2fa3cd2c0975e477a2de364", "identity.json": "8ec3c6c72438622d51772a37ebce2f3fb62aea0a9363ed75d7629ca06656ba65", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "31f9d06dd868b28f0f3b10250afb08f504aac34c692ff90e76b213ebde71d05f", "guard_own_pid_history.json": "02563d9f558a8b38d2d0405ea44dce7783b5cc6fb298d8024e72ffaeed981534" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_dit_rank0_10a11d3772070763_v1", "elapsed_seconds": 13.067716360092163, "allocated": "6", "input_sha256": "ea1184eaad12555ac80bf5ee33de7aa2251926b289572cadadb4043884848587", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 3, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "nvjet_sm90_tst_64x8_64x16_4x1_v_bz_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 4.672, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 5.803488, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 11.795973, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 15.37494, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.14034, "unit": "%" }, "dram__bytes_read.sum": { "value": 2.649344, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "byte" }, "dram__bytes.sum.per_second": { "value": 567.068493, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 165.332, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "78918143c985c5f0a71607d1da0764f35df19104eb4dd1e2c24b211916df758a", "metrics.csv": "cb20c331c27111df939e9eed97b745c16f21a1653c79efb6a0be063ed79a11ed", "operator.ncu-rep": "9bacd66af66c4a76c29eded9cffeee7ef38c86513043704c5781e5c0088cab7f", "run.log": "99c3c474962eae2e5ea001dd5869062b4755be61327464f2801130372a035717", "identity.json": "c87599360dc5390a6fee864f16a144232792f3a1e2ad63a818c6cf3ee2c10ec8", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "0bcd1e28f8ad31cea03d6dd56c27e8921320988066e2a708b65697da5e672845", "guard_own_pid_history.json": "e83ba4f33a6b13036258beb22fc81127a958e3512aa6662231d5eda0c9de8fdb" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_dit_rank0_ca590b3a668a9ecf_v1", "elapsed_seconds": 14.236177206039429, "allocated": "6", "input_sha256": "a6cb8d5375b991c9c5d1a1e5674fa86eb61d10648b2ccf71bb7ec40c56f2ac1e", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 3, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "nvjet_sm90_tst_64x8_64x16_4x1_v_bz_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 17.344, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 8.348938, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 66.960798, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 66.232482, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 13.346335, "unit": "%" }, "dram__bytes_read.sum": { "value": 52.487168, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 3.20896, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 3.211262, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 165.332, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "78918143c985c5f0a71607d1da0764f35df19104eb4dd1e2c24b211916df758a", "metrics.csv": "ff0ed187a9a695411c47f2a3e2bc64fa92dfad31300a626d8085c077e414338a", "operator.ncu-rep": "eb883faba257e91db4902c0eb3de28f64b3adc919f34579b9326162637e346ad", "run.log": "ba18f70cf2a53129cbfbe7160881ed0ec1b32922a6dad56c0ddb6c2a104c8571", "identity.json": "542003d571551940f94fdd884c200883e8335d7f265c6e1f454919aebf7ac712", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "1024b1e13d83f46facc0a637b215fcf59a0013500523934d79a7823f2b00efc7", "guard_own_pid_history.json": "f8e13421b73ae4ddc0b0781810214b4e749c412981750a484d8ca9ccb938b930" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_dit_rank0_22805ed4eb786e66_v1", "elapsed_seconds": 18.96791124343872, "allocated": "6", "input_sha256": "d6a2e382ecf0fdad711df2cd1bcb239a004321ae666d26aa04920718cf49bf81", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 3, 30720 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "nvjet_sm90_tst_64x8_64x16_2x1_v_bz_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 79.808, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 10.324124, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 83.319173, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 86.567175, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.452215, "unit": "%" }, "dram__bytes_read.sum": { "value": 314.74688, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 5.089536, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 4.007573, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 165.332, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "9caeb074fbd7a4b3310027643778504c64c82c15a7ad7a0dea80cada37441292", "metrics.csv": "461281d7eeb06dfcbe6015480ea440edec821aa42f46c491f276a01086fb0ce4", "operator.ncu-rep": "8fe6e68837a7d0acbb7b3367ea2caed028642f5c1c30f483dca49884d10e39c0", "run.log": "8b2398ffb9c29e8f22e745aafd6b13c1f29fabf0078f3f95e41985fd0c8586b2", "identity.json": "2c37572e7aa6e7b47306a78ac377a77cac7077b4ee5a972059319229f3e94c75", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "86a21f458b9ca84bc6435c03cbd4206765ffae149552e5ad7f1d8048499de33d", "guard_own_pid_history.json": "3214ed41ac617230700acc68d464340d4219e8383524ca95394bdcb21f74421b" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_dit_rank0_fcebe966800b94b1_v1", "elapsed_seconds": 21.532828330993652, "allocated": "6", "input_sha256": "4f1e9988bf86b3f91f274f9b8c3bf33bf42b39945181ce6f09fbce8e0132cb6f", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 1170, 15360 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 252.832, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 85.238963, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 24.405996, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 50.066814, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.6146, "unit": "%" }, "dram__bytes_read.sum": { "value": 262.954496, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 33.932544, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.174246, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 226.5, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "5562e1a5ce507a5dcc26d8e920a34c5fbfaea06754f92c8ac5cdcd96d0f34d36", "metrics.csv": "6ce786696e2bd2aa6bd27156d88c51f6bbad5cb8da821def44861f8877d32af8", "operator.ncu-rep": "fe316e09db9f4f566b5821c0774e7514b14a4dd33bca49e83696d7a08a4f94b0", "run.log": "72eb6dabc330f9cc5818ea00418465dbd79f3cc8e2d790c133879aa4085181d3", "identity.json": "36c201ed2b54cf0f7a205236dfd46de00bddc30fa4e0f20f35edd6a02b8303d8", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "d5d494bc6abb2f3cf9cd498b326fac47f6315bb5109d3a8a47436d7bb153b109", "guard_own_pid_history.json": "5213349faf310fcc22360f47b84c9a6de417e107fd662a7aae436022b6f5ce11" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_dit_rank0_10a9e2fea034c6cf_v1", "elapsed_seconds": 12.789992094039917, "allocated": "6", "input_sha256": "a64f7da6eb1fcc9779d40258d974eb529c2397aaba4df6e32363d7c5d4e73065", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 1170, 5120 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 87.104, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 81.160631, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 20.235254, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 42.130195, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.427094, "unit": "%" }, "dram__bytes_read.sum": { "value": 74.980864, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 9.79072, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 973.22263, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 227.516, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "df407afd8d725671f90fbf811564c700d59eb97ed29f61b450ac8c48854c470d", "metrics.csv": "02cea0e01e66f689def70badde18b4711e2c75e3d183fa6c857fa909b05eebd9", "operator.ncu-rep": "6dca379096c8738524b0a3871b25434a6ecf4a02250bead192c8852499f1e9ae", "run.log": "3ba8e1a82087e3387378fcf17f648b75dcfd10634aec26188c59a19d064a6d64", "identity.json": "286c88ab6916257d6683b64b54e5a1b87fcd73b02bc952967239431ac94ee2ef", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "a2208fe015188d54bb39330dea8681c69ae33ef9511235777164d082adf46c4c", "guard_own_pid_history.json": "276b60c936020451b9d4f13c7b2af12a8bb857b651de0705ff42b2bffe4a29e0" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_vae_rank0_6638f430ac58a438_v1", "elapsed_seconds": 12.298183679580688, "allocated": "6", "input_sha256": "32bcd9b4d8be26c876626649d4ff2e623424efe309222bcd7e80afc47310ded4", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 16, 3, 60, 104 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 2, "kernels": [ { "id": "0", "name": "void cutlass::Kernel2(Params)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 4.928, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 12.442371, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 5.21175, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 13.814379, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 7.051773, "unit": "%" }, "dram__bytes_read.sum": { "value": 1.225728, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "byte" }, "dram__bytes.sum.per_second": { "value": 248.727273, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 136.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 74.752, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 3.584, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 15.235932, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 7.119418, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 17.273848, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 46.267308, "unit": "%" }, "dram__bytes_read.sum": { "value": 1.212672, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "byte" }, "dram__bytes.sum.per_second": { "value": 338.357143, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "42be6bf151e12cc662957787638ad47b2148c9912919277e92ac365e63035dc0", "metrics.csv": "bf58bc7e296e1e73240163db03528d9e83724ea72a4549ba0113bd6442d39bc7", "operator.ncu-rep": "ce03f0bac6fcbb8dc367d0fde6e395f0485cee0a45510ecfc3a435e1fd839486", "run.log": "c0d27deb0f78f3cfce51f1cf4465adfabab06aa6fb8ecd875f0e65a23ed2baf6", "identity.json": "e04a2cbeb6d913cd591cb6033da8a4c8b5203dc9969d1df1885238ea28b52b28", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "853863405b16b176e8e62aee432647ae104cba8fc42e846d9b535783e2adaf6d", "guard_own_pid_history.json": "7bdc116a7f921ef70b1aced8b759d9c21d26c682beca141103a7a8ff5b1cc5a1" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_vae_rank0_29ef146661136bf4_v1", "elapsed_seconds": 15.955600023269653, "allocated": "6", "input_sha256": "2b9e80cee384305d79045df290612b04bcfe041bd67dc69a2bc096e45a6c5872", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 60, 26 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 4, "kernels": [ { "id": "0", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 2.816, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 6.541211, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 2.537262, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 7.217338, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 15.343409, "unit": "%" }, "dram__bytes_read.sum": { "value": 0.342528, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "byte" }, "dram__bytes.sum.per_second": { "value": 121.636364, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 3.232, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 17.657784, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 4.40506, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 11.185747, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 33.557148, "unit": "%" }, "dram__bytes_read.sum": { "value": 0.672768, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "byte" }, "dram__bytes.sum.per_second": { "value": 208.158416, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "2", "name": "sm80_xmma_fprop_implicit_gemm_tf32f32_tf32f32_f32_nhwckrsc_nchw_tilesize128x64x32_stage5_warpsize2x2x1_g1_tensor16x8x8_execute_kernel__5x_cudnn", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 16.32, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 20.669159, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 1.351313, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 16.424589, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 6.243267, "unit": "%" }, "dram__bytes_read.sum": { "value": 1.059328, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "byte" }, "dram__bytes.sum.per_second": { "value": 64.909804, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 123.904, "unit": "Kbyte/block" } } }, { "id": "3", "name": "void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 4.576, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 22.863568, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 11.044218, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 25.407107, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 71.799906, "unit": "%" }, "dram__bytes_read.sum": { "value": 2.412288, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "byte" }, "dram__bytes.sum.per_second": { "value": 527.160839, "unit": "Gbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "ccb1be2a78290c65e57f789dd00db0916eec43f608c21fad077988f7f471a62e", "metrics.csv": "3aba0e06fff40239486c3f155fa2503d25526a0cfaa39b92bb9f8e2060fd3134", "operator.ncu-rep": "f5185031b4ebeff1e09b053aedb775330bf2d760ce357dafd3d7d973d832a930", "run.log": "30ed758923d3d94f83cef325cab003415187dae14c3840f0e29ede3afe31b56d", "identity.json": "b57432c45b493528ac118148d4a5f4d3fb26076f24f0ecaa5c641e6bc7626c00", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "d63d5119f5cb3fef86c85a50ea4498be7f23f4c0f80b6a00591ac8d8ee8ba1da", "guard_own_pid_history.json": "bd1fd48aa894d8e87a56e4c5cd087b844b6cffe757e89e4ad43e064b5931e022" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_vae_rank0_551fa1ff3095f40c_v1", "elapsed_seconds": 18.08424973487854, "allocated": "6", "input_sha256": "6309c12f24013edabd88edced7c2ce22672322d203206b12ba218c88696effb5", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 60, 26 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 5, "kernels": [ { "id": "0", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 6.848, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 27.404928, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 24.416844, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 53.461993, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 73.664364, "unit": "%" }, "dram__bytes_read.sum": { "value": 8.008448, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.002048, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.169757, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 9.984, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 63.899285, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 34.966809, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 69.209084, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 84.724705, "unit": "%" }, "dram__bytes_read.sum": { "value": 15.93472, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.804352, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.67659, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "2", "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x64x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 97.824, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 37.579717, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 5.36294, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 41.273313, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 13.550294, "unit": "%" }, "dram__bytes_read.sum": { "value": 23.975168, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 1.262592, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.257991, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 232.448, "unit": "Kbyte/block" } } }, { "id": "3", "name": "void cudnn::nhwcToNchwKernel(cudnn::nhwc2nchw_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 3.648, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 14.880553, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 13.778107, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 34.779502, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 51.828874, "unit": "%" }, "dram__bytes_read.sum": { "value": 2.41152, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.661053, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "4", "name": "void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 4.544, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 23.116983, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 11.183862, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 25.764575, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 70.343863, "unit": "%" }, "dram__bytes_read.sum": { "value": 2.412288, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.530873, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "ccb1be2a78290c65e57f789dd00db0916eec43f608c21fad077988f7f471a62e", "metrics.csv": "12d48a0b8063b436147f604f8d269b1480e8e9947b99b78dfb9996f041d5da68", "operator.ncu-rep": "827ec5a15ba1542a0f43ee3c716c9d005bb5083c4326f520c9377dd95369ac40", "run.log": "98857b3d906c81ae105905c77352816a74710c0e48a1845aee1744e75a031cd6", "identity.json": "6e2c7ce18a4d432a58d93583d310b76684a8425e648c5e3f910a26a55e3684fb", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "7ed09fc3ed7a51ec955e194926674065759fa9012628a657d5effba946410fcd", "guard_own_pid_history.json": "94a2e0fb7ce614f14eae8f700b8ecd76b852c23fe6f43e69e8ec4aa9afca2409" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_vae_rank0_c16d11a7624e1a8c_v1", "elapsed_seconds": 17.119314908981323, "allocated": "6", "input_sha256": "1e43418295ca6422f27afcf0b9940aa818fbb12fcf1013c12dfcf7a60f86e7d7", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 120, 52 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 2, "kernels": [ { "id": "0", "name": "void cutlass::Kernel2(Params)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 12.384, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 25.920931, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 8.607037, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 33.983188, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 13.551422, "unit": "%" }, "dram__bytes_read.sum": { "value": 5.116928, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 3.072, "unit": "Kbyte" }, "dram__bytes.sum.per_second": { "value": 0.413437, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 74.752, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 8.768, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 46.554328, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 22.911033, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 51.457793, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 74.651981, "unit": "%" }, "dram__bytes_read.sum": { "value": 9.600512, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Kbyte" }, "dram__bytes.sum.per_second": { "value": 1.094949, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "d21e841f3d6a266873f218cb3c93c4c7fb7ab74b609ac99121d9419e29677cb7", "metrics.csv": "966631416bc3349cc1bd29b3a29b66c20b5642dd0523d5b553436567f63583c0", "operator.ncu-rep": "0246f668a98858dfada87af00835404adf45dd61a102e0292344ba5a3042f1e7", "run.log": "c4db0631a9a28afd16402a2736661c9044aa10532ed0626fd1e5db151a02d149", "identity.json": "90067b10cd34f960c77385f72954edd4f81112447060a8e84d324a71ae6e60fb", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "f8494164f3051281652eaf1bab61ff38b4daecaf282c0d7edcebc9aaec4b40a7", "guard_own_pid_history.json": "0cace774f89e5c20679bd5216b3e72e070fa1c6d33dd5e92b6d5ef0af168aa12" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_vae_rank0_081e405bd1e07dce_v1", "elapsed_seconds": 20.28731870651245, "allocated": "6", "input_sha256": "3193a32ddc024e9d34eb5446f9909408dde38d963ebc5cb9654998ecf661a820", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 120, 52 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 6, "kernels": [ { "id": "0", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 9.568, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 36.450718, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 34.985392, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 70.74527, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 82.59028, "unit": "%" }, "dram__bytes_read.sum": { "value": 15.187968, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.875264, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.678849, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 6.624, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 48.401978, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 25.123701, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 55.958628, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 80.361473, "unit": "%" }, "dram__bytes_read.sum": { "value": 7.971584, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.000768, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.203556, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "2", "name": "void cask_plugin__5x_cudnn::init_device_workspace_kernel>(T1, bool)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 3.68, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.028338, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.026364, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.302623, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 1.519546, "unit": "%" }, "dram__bytes_read.sum": { "value": 0.004608, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.001252, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 16.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } }, { "id": "3", "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_on_kernel__5x_cudnn", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 122.688, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 56.462241, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 5.56358, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 51.396564, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.433255, "unit": "%" }, "dram__bytes_read.sum": { "value": 24.830208, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 8.004352, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.267626, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 232.448, "unit": "Kbyte/block" } } }, { "id": "4", "name": "void cudnn::nhwcToNchwKernel(cudnn::nhwc2nchw_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 7.2, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 27.470815, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 27.859537, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 58.099129, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 81.932699, "unit": "%" }, "dram__bytes_read.sum": { "value": 9.6, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.018176, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.335858, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "5", "name": "void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 8.8, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 46.141705, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 22.736937, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 50.112379, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 73.619081, "unit": "%" }, "dram__bytes_read.sum": { "value": 9.600512, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.090967, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "d21e841f3d6a266873f218cb3c93c4c7fb7ab74b609ac99121d9419e29677cb7", "metrics.csv": "e65b3a2773150323d43a62874e0ab021c96847e6456ffed465200382f326d5a5", "operator.ncu-rep": "0045a9daa7de72e129a06a8835a25859e960ce5e81668d17c0ceb354969ec5a7", "run.log": "5289664cc0a910b98384d742a9f98046c015aed395ac83517a45f7a075f77f08", "identity.json": "a86741e54e573d3590f780044bb07445ac7a57b3cc7770ee3b707d9e85b82fb5", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "caa37c0390891ff86d7c93df8f8c6725dc32f18695d55f368c20d5d7f30b9475", "guard_own_pid_history.json": "a5797521e7b0576f48d70ffa72ee5e52036bcea115aaa254788b8ee69c014b57" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_vae_rank0_8601bc3a88374d99_v1", "elapsed_seconds": 19.31157684326172, "allocated": "6", "input_sha256": "c7f10fe8b8ca84229e785645c70c003456c3950c79cdd176694e5bc86c9247f4", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 1, 120, 52 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 6, "kernels": [ { "id": "0", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 16.64, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 41.695108, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 52.385208, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 79.417837, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 86.940903, "unit": "%" }, "dram__bytes_read.sum": { "value": 30.366976, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 11.497472, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.515892, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 10.144, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 62.70737, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 34.251154, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 69.326632, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 83.711812, "unit": "%" }, "dram__bytes_read.sum": { "value": 15.934464, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.766976, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.646435, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "2", "name": "void cask_plugin__5x_cudnn::init_device_workspace_kernel>(T1, bool)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 3.776, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.027533, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.027022, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.303405, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 1.552996, "unit": "%" }, "dram__bytes_read.sum": { "value": 0.004864, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.001288, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 16.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } }, { "id": "3", "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_on_kernel__5x_cudnn", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 240.352, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 57.441182, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 7.155502, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 52.670803, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 13.956357, "unit": "%" }, "dram__bytes_read.sum": { "value": 68.61952, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 14.121216, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.344248, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 232.448, "unit": "Kbyte/block" } } }, { "id": "4", "name": "void cudnn::nhwcToNchwKernel(cudnn::nhwc2nchw_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 7.392, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 26.789002, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 27.118987, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 57.452752, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 80.668169, "unit": "%" }, "dram__bytes_read.sum": { "value": 9.6, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.006144, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.299532, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "5", "name": "void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 8.8, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 46.241571, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 22.771544, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 50.164747, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 70.432619, "unit": "%" }, "dram__bytes_read.sum": { "value": 9.600512, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.090967, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "d21e841f3d6a266873f218cb3c93c4c7fb7ab74b609ac99121d9419e29677cb7", "metrics.csv": "f31995cc8246a57dfaac0d7653269ccfbd0f244a053a5a7a0ee54e5e41aacedb", "operator.ncu-rep": "5dd349e2a57692f5a1a7b75d2c63ec8c8d4718cc78403b2ccfe711d9a1f3eb4c", "run.log": "2f273665a8b885ad53cfd0552f5fad2012a7c0791f95d2113296b219b38efe25", "identity.json": "046aeec6c638f9749bfc3b1052c616036de393c2ebf30c25e77403b7d32012c6", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "7103347f204bad1f0a8edfd6dd6348278a78594ded080ebe01e779bebca91a98", "guard_own_pid_history.json": "09b74dfd830d6b2d9be3e0b911ea7860166f066459d958c9071dda49035ef63b" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_vae_rank0_99475e040d65d084_v1", "elapsed_seconds": 19.041394233703613, "allocated": "6", "input_sha256": "0cc0a21ce474602c53c3e5621e64598ed7be5a2c16603fc968441df41de71330", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 192, 1, 240, 104 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 5, "kernels": [ { "id": "0", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 34.08, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 39.224356, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 60.726759, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 81.305041, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 90.207582, "unit": "%" }, "dram__bytes_read.sum": { "value": 59.112448, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 40.355328, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.918655, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 4.896, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 33.361319, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 17.081921, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 40.884687, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 73.72203, "unit": "%" }, "dram__bytes_read.sum": { "value": 3.990016, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.814954, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "2", "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 166.176, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 85.708077, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 9.421867, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 68.827558, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.85885, "unit": "%" }, "dram__bytes_read.sum": { "value": 64.303104, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 11.027968, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.453321, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 232.448, "unit": "Kbyte/block" } } }, { "id": "3", "name": "void cudnn::nhwcToNchwKernel(cudnn::nhwc2nchw_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 10.752, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 37.000924, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 41.05373, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 73.968927, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 83.475165, "unit": "%" }, "dram__bytes_read.sum": { "value": 19.1872, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 1.970688, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.96781, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "4", "name": "void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 14.816, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 54.648822, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 27.255233, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 58.75947, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 77.381408, "unit": "%" }, "dram__bytes_read.sum": { "value": 19.18464, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.17664, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.306782, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "898fd24576efad478842a83a33c606603acb1c4923ddc647361a10700df64ca7", "metrics.csv": "8e04dd103a921257d405955a2acf864a485740280690469b34a8aabc8043b9a3", "operator.ncu-rep": "46cf2c00e51730fc3514e2c8580afc2f7a51273eed286560efaeb58dd428e17f", "run.log": "8f2f79c5dbcfdb23b623a8ee77def4f6153c5c5d99962514691d68696bfbaa69", "identity.json": "850d027e34b87f2d3aa4528a40aa08d48cc6b138eeae2b87b06906b9f31f4fb2", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "e08bac240f375ae35eb425ab0515905a03903f23f79fc8cdd110d09a9ac5e7cd", "guard_own_pid_history.json": "e51fb0e05e5c17dcc9d8bd122554a244bf3df37559a9fa459b255d82785c3d57" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_vae_v1", "elapsed_seconds": 16.918233633041382, "allocated": "7", "input_sha256": "1e47965836d719f164296b5c8d52cac8c22a10d9ff8c5f4fdfa48498909f8856", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 96, 1, 480, 208 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 5, "kernels": [ { "id": "0", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 70.336, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 37.540406, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 63.710272, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 80.815686, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 93.308252, "unit": "%" }, "dram__bytes_read.sum": { "value": 116.620544, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 98.860032, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 3.063589, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 3.52, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 12.390404, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 5.953501, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 15.414774, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 24.41718, "unit": "%" }, "dram__bytes_read.sum": { "value": 1.004288, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.285309, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "2", "name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 345.6, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 29.68631, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 8.985101, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 60.273264, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 17.755377, "unit": "%" }, "dram__bytes_read.sum": { "value": 117.670144, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 31.733248, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.432301, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 232.448, "unit": "Kbyte/block" } } }, { "id": "3", "name": "void cudnn::nhwcToNchwKernel(cudnn::nhwc2nchw_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 19.904, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 41.754029, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 58.781261, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 81.369093, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 89.587993, "unit": "%" }, "dram__bytes_read.sum": { "value": 38.355968, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 17.823744, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.822534, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.248, "unit": "Kbyte/block" } } }, { "id": "4", "name": "void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 27.264, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 59.158799, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 39.056992, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 60.717071, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 79.24734, "unit": "%" }, "dram__bytes_read.sum": { "value": 38.354688, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 12.763136, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.87492, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "fbc0f1cc0275724c75a7685d0c3da84472740f5dee6ac7697f4160715261d45b", "metrics.csv": "9715aa5feeaf6b673c12be1eba462cb072e0fca573ef8dd4cb696fb4ef5e2a96", "operator.ncu-rep": "c9a1af7a6b873a50b0bf97afac127c0a95a95393722b41b707f51989c697a292", "run.log": "320072147bf1c95d44cd928524260654a333c7576eaa37dfeacd0a891ad9bc4a", "identity.json": "e5ea23ca715cea4c1ff59609f629ec7157e8de7ab3cc28bde2a26b0d351570a3", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "d3bbc32df63026796e323fc0bdf4665c84eb041651437c3f366df91fc662f20e", "guard_own_pid_history.json": "b605d0611a41d8942514271cc07d3ff302a799289f74f359744d9e89a9ce5ec5" } }, { "stage": "output", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_batch_output_rank0_f18eb91117dda1c8_v1", "elapsed_seconds": 12.896703720092773, "allocated": "6", "input_sha256": "897dd2bd3ad5fe7258ee5c3ae9cfd93da4e5e94644902267f0b3d8bd156f63b0", "correctness": { "op": "aten.to.dtype", "stage": "output", "shape": [ 1, 3, 9, 480, 832 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "void unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, LoadWithCast<1>, StoreWithCast<1>>(int, T1, T2, T4, T5, T6, T7)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 29.824, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 55.696, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 33.813727, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 43.414546, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 86.444578, "unit": "%" }, "dram__bytes_read.sum": { "value": 43.166464, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 5.343232, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.626532, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "445e5fabf094c7c0c3c9f40b29392e37a47ebfbac59a0b0a69e8f82c9d3e0073", "metrics.csv": "21a33e9eac00633e4c8000dabf6158110f2d0b379b1006144b11779474998d09", "operator.ncu-rep": "339aca9bdbfe96e2e0974daa43e41a127f20157e53acff5ba12fa0cbc9eb1451", "run.log": "1c386c58a42c31bf1cc7f3ad9b61b20199890afc03d3168a1e2a47e6b09e5164", "identity.json": "03d01b7b12ac8930ec576e0125d9592068f4abbb0841cf1136be0c8a4b66ef0a", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "21f0c7959fec28ca3be30ef716878ab682a86b07c113d34ecdc75b961be88e55", "guard_own_pid_history.json": "215db1193797a4181bef6b32e0f47904fbd50bcc87c7cbc001fdc73af7ebecdc" } }, { "stage": "output", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_capability_output_v1", "elapsed_seconds": 11.483896017074585, "allocated": "5", "input_sha256": "a9aab61b63e7cab483e81582c9af41d89d525814ab14de753e8a7b16de6c4f1b", "correctness": { "op": "aten.to.dtype", "stage": "output", "shape": [ 1, 3, 12, 480, 832 ], "exact": true, "max_abs": 0.0, "rmse": 0.0, "finite": true, "math_policy": { "matmul_precision": "highest", "matmul_allow_tf32": false, "cudnn_allow_tf32": true, "cudnn_benchmark": false, "cudnn_deterministic": false }, "scope": "Original ATen op on saved real tensors; isolated diagnostic" }, "kernel_count": 1, "kernels": [ { "id": "0", "name": "void unrolled_elementwise_kernel, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, LoadWithCast<1>, StoreWithCast<1>>(int, T1, T2, T4, T5, T6, T7)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 39.584, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 55.888331, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 36.344303, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 44.802352, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 89.899207, "unit": "%" }, "dram__bytes_read.sum": { "value": 57.544448, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 11.594752, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.746645, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 32.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "59ec25fab08d094e717140a7b0928a857ff389dbec068b03d6d3100ac8242bbb", "metrics.csv": "ef8d7d820d04ff1bb9f5f0e42b670acbf7ea11499b7e808a05a41c42d6b8f866", "operator.ncu-rep": "c82f979792422d72988256c4fd4707e899701d0a595c81f9689df3fbd3cb3c9f", "run.log": "eded896000896d7f5bd8b96428023fe17082d40a53629f054ead5b4381376ed5", "identity.json": "a9ba2df59f68800ced66deaaafc40607d26cbc1ecab3320c91dace7266f99fcd", "replay_operator.py": "21e6ebbe13e53933591ad14feaa6ab4b189738f3647f2087ce8bf523b0e07450", "manifest.json": "4bb8b7cfbf7552414ab9e7358a851a245455039059d9a25bf68a2ab932203d9d", "guard_own_pid_history.json": "3e842381e209236ec558c2bbe27aa6d52e9cf1b4135e0b08afa6b84bb2d326c9" } } ], "scope": "Eighteen isolated real-input operator replays, single device per job and distinct allocated devices across jobs. Exact reference output passed. NCU replay instrumentation and clocks/cache effects prevent treating these timings as clean integrated performance. No whole-model MFU or optimization acceptance. Guard history is a point-in-time check of each own PID." }