{ "generated_at": "2026-09-10T04:43:00.333974+00:00", "status": "passed", "jobs": [ { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_80ac9ebe75667de4_v1", "elapsed_seconds": 13.737869262695312, "allocated": "6", "input_sha256": "b8c205d5140f4a6d1aabaa7d0946d8eb3aa5bdc6007bff3e8efd26fcf1740c87", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 1170, 13824 ], "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": 215.52, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 91.537864, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 22.141116, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 53.502631, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.666908, "unit": "%" }, "dram__bytes_read.sum": { "value": 199.619072, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 29.967872, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.06527, "unit": "Tbyte/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": "82612d9f55bfb23ec738122e2e6fc9ba06db45994725b28a47a22d7463a51781", "metrics.csv": "622ae0f78b6fa48ef97e64ccfee2d66b43a3c7e23e6daa664771165b7eaa1290", "replay_comparison.output.pt": "4814de55f45e584a68066a4fa21bbb1c3fd43a1e40c2f3de5965ddbcefc06556", "operator.ncu-rep": "a7c97b10e527a373da02e9d900f722e17794371d253dd86ce293aed0049cc468", "run.log": "fd4c22ab47a3a232609398abc2be7e4afdfac8e375c9cc87fbe9391e85b9febb", "identity.json": "25f867b9bac59c12a3108010d0f349082eae0d8dcea3cd3ac71d6c59d9598097", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "9ba39aed4c4af8fd8db5d489fe1762779de2f7196970cc7eb11cf1b8989b2f2e", "guard_own_pid_history.json": "39f41fd24cf9be86d65a393dbf2964c2ff063d1fa2ed18090f616732497ae865" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_ef1fe591ff44862b_v1", "elapsed_seconds": 13.203153371810913, "allocated": "6", "input_sha256": "d89811d1f0f1ab9382656380f7ff354f19e521d815cd71b6e39b7d2a72def4ec", "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": 220.608, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 85.960958, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 20.442762, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 46.922333, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.415541, "unit": "%" }, "dram__bytes_read.sum": { "value": 206.323712, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 10.662912, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 983.584566, "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": "7288e1b9653aef916cb445b78fc52edb136cb42f7c4fa49aaf8bb7effa0a5b72", "replay_comparison.output.pt": "793c5c04f09b4b561b8161ad42857f18d0c505c25d62202a32ce706d19b17e25", "operator.ncu-rep": "2cf6d7529ebf41b7e4eac2eea784083c9cb9b12684d6c4c6dcb9f116c56b02ae", "run.log": "aeb067a29085327a0c829880e201fcc4bb171ac2f8e93fc1ecf9605aee79d00a", "identity.json": "2c25c20527043c19317a7085321a15da445acc5e5554688a98702bf4f4e30149", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "2e12ed660b4763de51611f8da2507a60807d0a02587d7282f9aae48f6397f71e", "guard_own_pid_history.json": "885134acacc8018a4a3a3422c3e287d53ef6c2dd959d711cdc63dea6d9af1824" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_6d01e6f8a09a63ac_v1", "elapsed_seconds": 13.180827379226685, "allocated": "6", "input_sha256": "7de91ad08752fccf4380732c881403df8972bc20e43941ce68d7ee08b8d5d769", "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": 85.696, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 81.017145, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 20.578043, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 44.816086, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.417252, "unit": "%" }, "dram__bytes_read.sum": { "value": 74.979584, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 9.85472, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 989.944735, "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": "b48f09f385e3618afcf0cd2a3bd14eab33684a19bb27cf1a4514188def9dbadc", "replay_comparison.output.pt": "81b042f19e0af5a93a41103fd866d2eb2d3ed1508da038008b138f896fde23b2", "operator.ncu-rep": "3c6395230c151323ae2279b5c3743942a6168a6ef426b393b1dea85f8512da90", "run.log": "1a16e10485227470a364806f0a6c05242838fe012b273b90273792b784c8f41c", "identity.json": "51827d83677a09a7a936ce22d7163281e22fe4711814493445507b90302803a0", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "95e187f03c4ceeaae9bb72d30f9d5eff265548d076f069595e183db73bfbc33e", "guard_own_pid_history.json": "748d842aa20bb8e944c050107bc1177f1f28bae7c2dd27df9c7e05d485f17a86" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_41fea95495c28b56_v1", "elapsed_seconds": 18.662790536880493, "allocated": "6", "input_sha256": "7ea1828a641121c7e50c8fec13aa00d78a4f4b0f419741b8b6e74c3fd8f2b955", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 192, 4, 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": 68.352, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 39.125069, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 66.195507, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 82.796731, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 92.678842, "unit": "%" }, "dram__bytes_read.sum": { "value": 118.218752, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 99.319552, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 3.182618, "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": 5.216, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 31.247078, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 16.025307, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 38.455553, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 74.300598, "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": 0.764957, "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": 706.24, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 78.229995, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 9.312729, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 68.377794, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 18.165615, "unit": "%" }, "dram__bytes_read.sum": { "value": 244.80384, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 71.66336, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.448101, "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": 39.008, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 40.024653, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 70.528428, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 85.497306, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 90.890535, "unit": "%" }, "dram__bytes_read.sum": { "value": 76.694528, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 55.604736, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 3.391593, "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": 53.6, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 59.654443, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 49.625097, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 62.720143, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 81.25356, "unit": "%" }, "dram__bytes_read.sum": { "value": 76.694784, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 51.2256, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.386574, "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": "ab791b6b06c9a22ba54c97b2307358a1deb1537686fb0cbd6fb5b544f5b96be6", "metrics.csv": "c94cb12a6416d045456155d33b733aa1da369b185e3a198c1660a54c127f1ddf", "replay_comparison.output.pt": "0bb412c205f3bc08eb9903b194f8220422aa0a7408176e3ef85a92da6f436210", "operator.ncu-rep": "dcb4dbb2c59e017b7d4a655b2c049a01b160c59aba942dc185d8fadc74281247", "run.log": "b70b56215527006eb764788950c2389bc98e31852d63d0b91ad698f2f1611552", "identity.json": "24e58867021c888cbc95d312535e711b12ac687129eab608ead6a913e6788f30", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "b67870ca1946901002f7a0b958d1de0b492a74610b21c117a2ddb13fe704eb71", "guard_own_pid_history.json": "3c528d5f3442a7749bfdcc1f2f2def1b7b7863814caa98d2a06c9ada18d43222" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_74aff8f080d200c8_v1", "elapsed_seconds": 18.323726415634155, "allocated": "6", "input_sha256": "3da5719db48ceca956fb363d6dbfa423964bbde96b9ce8b8dd9399261d9405da", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 96, 4, 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": 0.140096, "unit": "ms" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 38.200711, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 66.546526, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 81.38646, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 92.912553, "unit": "%" }, "dram__bytes_read.sum": { "value": 233.226496, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 215.286784, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 3.201471, "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": 0.003648, "unit": "ms" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 11.926683, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 5.743217, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 14.876259, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 25.684751, "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": 0.275298, "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": 1.34, "unit": "ms" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 30.613135, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 9.524368, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 59.103268, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 18.22425, "unit": "%" }, "dram__bytes_read.sum": { "value": 467.56096, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 146.57536, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.458311, "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": 0.08096, "unit": "ms" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 41.056031, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 73.516066, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 84.811005, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 93.288254, "unit": "%" }, "dram__bytes_read.sum": { "value": 153.371136, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 132.944384, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 3.536506, "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": 0.107456, "unit": "ms" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 60.133901, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 54.32714, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 63.797536, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 83.574473, "unit": "%" }, "dram__bytes_read.sum": { "value": 153.37088, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 127.412736, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.61301, "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": "d90423fe4a4f7daa5f8fccac8fe5ca626afaed80deb319e9f96f9b7ecc8db54d", "metrics.csv": "e88a2c4abf33df6793433b2805a6160ca9ad2a1b98f30016b5ec378239163b93", "replay_comparison.output.pt": "1e95ef6cf2d60bfafe89cc296461d6986f8301752c00295dccee0dc71842ee06", "operator.ncu-rep": "8ad01a999ad2d69c1751c0cff33d5d3b03b8fc186fc1818a8515ccbedbe499ab", "run.log": "cb8dc87519311f1e39643280bbf570b1cf48a9c372a0aea94440de4ff5bbbbac", "identity.json": "05a44e64aa1d24e909d2c4a684ab901fdd81d1d0b8741fb1c961811ad107aa97", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "61a387d99db2a0799d3b97f2ec9951a20976f3d08c9c6449f29441a1a559a16f", "guard_own_pid_history.json": "c1135888130558d6e3e7dad50b43152c21876e14f1c6e8be305944097dff46a5" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_7fc47a1ee259ec1d_v1", "elapsed_seconds": 18.977525234222412, "allocated": "6", "input_sha256": "538cab189b8721f3989e60d331a5e8f3a0b3ea6e3f542b4f5be45a3f1807f370", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 2, 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": 5, "kernels": [ { "id": "0", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 22.304, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 41.271243, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 56.938789, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 79.357066, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 88.712471, "unit": "%" }, "dram__bytes_read.sum": { "value": 40.485888, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 20.514304, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.734944, "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": 10.304, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 61.769375, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 33.529028, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 67.254028, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 84.865132, "unit": "%" }, "dram__bytes_read.sum": { "value": 15.934464, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.635392, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.608099, "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": 427.488, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 65.894637, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 6.41846, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 53.429551, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.878182, "unit": "%" }, "dram__bytes_read.sum": { "value": 113.32224, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 18.706944, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.308849, "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": 11.2, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 34.827828, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 39.394113, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 71.450544, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 86.668335, "unit": "%" }, "dram__bytes_read.sum": { "value": 19.185664, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 1.973248, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.889189, "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": 14.944, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 54.104812, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 26.885869, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 57.817172, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 75.631676, "unit": "%" }, "dram__bytes_read.sum": { "value": 19.185408, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.086016, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.289576, "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": "31b28e8acac4d172e8229285efd6b166430ed8b99750bb4d697e38d1e768012a", "metrics.csv": "c56332cf4b9d241312b9b44c5073a49d50f7891aedc91607ce1ff55ae26f221b", "replay_comparison.output.pt": "ef660542b3c107b46bd33003204d5df07c9783083efe46d0ac703196b5009b9c", "operator.ncu-rep": "98916fe5bb8ef34404657b7e06e076374f89fe0834721326c58e1cf561eb5d3d", "run.log": "db48d9c7e89b77ded6e888acd069880b5fd9ad19a782bc3094d8bbe68c22c160", "identity.json": "47c84e8841dbe8cd0e6a6bdfe91b306ca25ef9fb15acdfd219420011240f0994", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "55609465fed33160a0d31558af0f2c90c750fbad334fee14eb7a27692d3af32b", "guard_own_pid_history.json": "d25709778c23a886dacc3e9706c56a30089b509af8388a3c71a63a6e12490891" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_e49e1cfe69d26b07_v1", "elapsed_seconds": 19.300371646881104, "allocated": "6", "input_sha256": "6fa08b3e20b8b3a0f9f34750ff5afbc4a19f18369bd70ccbf0ed0d1071ca5f08", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 2, 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": 5, "kernels": [ { "id": "0", "name": "void cudnn::nchwToNhwcKernel(cudnn::nchw2nhwc_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 11.456, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 40.528654, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 42.349603, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 75.323444, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 85.832553, "unit": "%" }, "dram__bytes_read.sum": { "value": 20.249088, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 3.02208, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.031352, "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": 6.656, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 48.175667, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 25.012249, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 54.92529, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 81.48079, "unit": "%" }, "dram__bytes_read.sum": { "value": 7.971328, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.000768, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.197731, "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": 209.216, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 68.580954, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 5.12333, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 52.419661, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 14.794025, "unit": "%" }, "dram__bytes_read.sum": { "value": 38.672128, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 12.901632, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.24651, "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": 11.136, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 35.060846, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 40.414344, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 72.631446, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 85.495716, "unit": "%" }, "dram__bytes_read.sum": { "value": 19.185664, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 2.391552, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.937609, "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": 14.56, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 55.602228, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 27.777187, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 59.956106, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 77.608876, "unit": "%" }, "dram__bytes_read.sum": { "value": 19.185152, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.22272, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.332958, "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": "31b28e8acac4d172e8229285efd6b166430ed8b99750bb4d697e38d1e768012a", "metrics.csv": "430a85d9d1d993854321e6bd2ab791581ce2262a73d999e2ead2d5ef446254af", "replay_comparison.output.pt": "3fab95118d2eee216ec8f735edfcd84964025c36f99ad240b1a59946ee35eb7f", "operator.ncu-rep": "631501c0a8a2597a9ef5d02ddbd11b52ba387575d7e942e0e67c37e12928e945", "run.log": "c4319ea04ec02a911489115f5a6ba67ae2942ee3bea1f5d63f328bdfc07dc186", "identity.json": "4d35d82454182312d77e7b220ecb479670ad3d03e438ece0b710f4372ef27351", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "796c1208b7e9c2d5d54df901972e50cfe125570f8b71dc201aa95f91c6751299", "guard_own_pid_history.json": "9162d5e83ef4f1a12d9bb06e1167c43099b31e0e88eb70b215fb796929200fef" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_c87dc674a51efcf2_v1", "elapsed_seconds": 21.19285535812378, "allocated": "6", "input_sha256": "3547fbf5eb18cf7da7abff35f3a785e4dd1cfc51a243c23698a162ff25086677", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 768, 2, 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": 20.224, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 43.18329, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 58.463134, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 81.060841, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 91.642053, "unit": "%" }, "dram__bytes_read.sum": { "value": 38.349568, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 18.360832, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.804114, "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": 15.2, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 82.620515, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 4.858398, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 12.826662, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 86.569261, "unit": "%" }, "dram__bytes_read.sum": { "value": 3.547904, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.233415, "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": 3.52, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.027775, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.02746, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.311923, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 1.554039, "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": 0.001309, "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": 81.536, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 60.214853, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 23.653585, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 58.177611, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 16.859649, "unit": "%" }, "dram__bytes_read.sum": { "value": 59.979008, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 32.811008, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.138025, "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": 20.032, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 38.181248, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 57.907027, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 81.941375, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 88.130031, "unit": "%" }, "dram__bytes_read.sum": { "value": 38.358784, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 17.398784, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.783425, "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": 27.712, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 57.998715, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 38.31248, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 59.274381, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 80.19311, "unit": "%" }, "dram__bytes_read.sum": { "value": 38.356224, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 12.673792, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.841441, "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": "100a02f11f3eceb743ba023e84c4812ff66e0d99cb9d07cc88c5e2945f7cbd41", "metrics.csv": "8b7ac04408b03cb6445e5e687b2ec244e5bb81c30e766768b69d3f57ae977ef5", "replay_comparison.output.pt": "9f2e8129d46a9ddf6d3d4368c3c862a8b79595c782268f37336972711ab80825", "operator.ncu-rep": "24eda23decc9e412cd6a5ecb92ad2145f2856225a48c19f8ceec059c1ec25bd9", "run.log": "00b2fb25ef7281caeaee53b8074e840a427566cc9fc592618598eae3329bbfff", "identity.json": "85686ae68de947785d70e66588b6b3161333811cecc25b5c2fcdd31ed6878f1c", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "2ddad0d2ed89c88c760ae7f53a19e1618fc0d65b8ca5de2160e323189f4de83d", "guard_own_pid_history.json": "34646db5cfeeb74ad915e4d8e832c44635e84195b784360f91b8cb8aeb8dec6f" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_27b8a5265a151590_v1", "elapsed_seconds": 21.660016775131226, "allocated": "6", "input_sha256": "3558a52a268a76189298b60a739c4bc3a01099db809b69c10633a6c964ffd111", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 3, 4, 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": 139.52, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 38.236556, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 66.678149, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 82.169336, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 93.600613, "unit": "%" }, "dram__bytes_read.sum": { "value": 233.225728, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 214.398976, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 3.208319, "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": 3.2, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.366427, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.266516, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.908733, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 12.423183, "unit": "%" }, "dram__bytes_read.sum": { "value": 0.040704, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.01272, "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": "sm80_xmma_fprop_implicit_gemm_indexed_wo_smem_tf32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x16x64_stage1_warpsize4x1x1_g1_tensor16x8x8_alignc4_execute_kernel__5x_cudnn", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 536.64, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 49.269089, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 20.583373, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 69.006466, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 29.756278, "unit": "%" }, "dram__bytes_read.sum": { "value": 523.764736, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 7.751424, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.990452, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 94.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.12, "unit": "Kbyte/block" } } }, { "id": "3", "name": "void cudnn::nhwcToNchwKernel(cudnn::nhwc2nchw_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 18.784, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 74.116556, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 5.334298, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 13.932529, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 86.846153, "unit": "%" }, "dram__bytes_read.sum": { "value": 4.807424, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.000256, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.255945, "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": 5.984, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 34.462223, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 16.846614, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 38.220111, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 68.534774, "unit": "%" }, "dram__bytes_read.sum": { "value": 4.806656, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.803251, "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": "5be3369f6e18bc5e4c9f74e68dbd8a202c415f5c7cd56e5422c153e0bd6512de", "metrics.csv": "aa5929288bfd8257714b295ee3f0e82d7c093f32a9919b1378d00c535256eab9", "replay_comparison.output.pt": "d1ff4e01d66be83476061ff782801ac27c7ab9a1df9c61e6b17710cf18f3fc1a", "operator.ncu-rep": "3664d715d416fbbd4f9f7c0228e13e11e04ae1f74c2f3b08aa1437bef7194a61", "run.log": "3e8bb224b39a867b1fdcd488290d7b8176a0bfdd319442dce5004873a57478da", "identity.json": "a1e1f04589492f00d92e7cb88007c38563d48f78bdaf056284d7a6b468a7be3e", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "e3c11b2506a3d321dccea15d34d2abb9c7fd9092bbc668e05f240af8531bc7b4", "guard_own_pid_history.json": "7917008b67306d75527c0d8c987ff4451b1240bd824c3330c452dbd0f3532a33" } }, { "stage": "dit", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_301783745952cfe0_v1", "elapsed_seconds": 14.90436315536499, "allocated": "6", "input_sha256": "34457f152c46e24a2e317b09c8ddc97428430b2e02f99cc2ddeb74c8c157ed9e", "correctness": { "op": "aten.linear.default", "stage": "dit", "shape": [ 1, 4680, 64 ], "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_64x48_64x15_1x4_h_bz_bias_TNT", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 17.6, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 18.383383, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 60.982457, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 62.360801, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 13.831379, "unit": "%" }, "dram__bytes_read.sum": { "value": 48.598016, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 3.023104, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 2.933018, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 168.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 232.892, "unit": "Kbyte/block" } } } ], "sha256": { "replay_comparison.json": "0140faadfb462ea2e3b5516c62b4b174b29b724e054d0412adf73e5b85cb5d54", "metrics.csv": "08ca6ae18d06e5584d4f754370f2141d95de2694d3c930c640c2ce62701483fd", "replay_comparison.output.pt": "0d9539a278261d4b6b417304f455d57690f884b1125516643288752452b4fbeb", "operator.ncu-rep": "bad694d7a2181aa808ee0eae5551790a941b6522ffc131414152ee56ef6ca35f", "run.log": "1ab9c46e07eba9d10bce6d1c576c7e47cbdb407dbd90ce5fb4fc8229b3a698bf", "identity.json": "b1c2de0e0a0933200af16ebb26af4955dd2a8290d8f868a1786f3ae98f5424b9", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "216e29db069ee92d6a0120b7c9f9fac1fac13ab27d681ce73998801eff445b96", "guard_own_pid_history.json": "a753f6c895e38dca0920ffffbde3361d4d46e13d6c1b1518ee25b5fcfedcca06" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_d35bb120bedeb7d0_v1", "elapsed_seconds": 23.974581480026245, "allocated": "6", "input_sha256": "03221b37d95f7c65a91055eae771f796fbf8e0d7758ce6840701ea0c066358ea", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 768, 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": 6.144, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 27.570005, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 24.455907, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 53.828489, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 68.387697, "unit": "%" }, "dram__bytes_read.sum": { "value": 7.19744, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Kbyte" }, "dram__bytes.sum.per_second": { "value": 1.171458, "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": 15.328, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 82.067119, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 4.830805, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 13.075698, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 84.833643, "unit": "%" }, "dram__bytes_read.sum": { "value": 3.547904, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Kbyte" }, "dram__bytes.sum.per_second": { "value": 0.231466, "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": 18.688, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 32.031099, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 12.041761, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 31.760624, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 11.61247, "unit": "%" }, "dram__bytes_read.sum": { "value": 10.793216, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 1.792, "unit": "Kbyte" }, "dram__bytes.sum.per_second": { "value": 0.577644, "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": 5.376, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 18.836147, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 18.627188, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 42.822127, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 75.330003, "unit": "%" }, "dram__bytes_read.sum": { "value": 4.807424, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Kbyte" }, "dram__bytes.sum.per_second": { "value": 0.894238, "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": 6.176, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 33.110637, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 16.232188, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 37.095517, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 69.871775, "unit": "%" }, "dram__bytes_read.sum": { "value": 4.809728, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Kbyte" }, "dram__bytes.sum.per_second": { "value": 0.778777, "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": "1ece88c4fa6389b5d316a722e3c6bc00bf77dba00a2e6e83cb41e083ca9696e0", "metrics.csv": "f40bc37db6e1b54e537917e6ff1078ae1a57d0bb921ebf469b6fb5130244a517", "replay_comparison.output.pt": "669c011e7179a173b0f7d96fe9296ce52bd74598b9b0d159cea206728e4733c0", "operator.ncu-rep": "86a4b60334b653bf6533a5bf89b13b4b5f2f9d2784b63e29929bade5c4f98767", "run.log": "6d0817ce9848ebbb1f4611a91d536d5bdc95728f4286e5c106b392e94d480b77", "identity.json": "508dc3426b1c69f7a829d9425891f593369a2d88fcb095e987ac61a4839910c6", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "afb27457b238f03ccd09aea5e3c7521abde21c2bd64fccabda5ecde2b6840956", "guard_own_pid_history.json": "28dbd799f51e8ed866733829039392a64c8b8b72341df073b00dfde517973dc5" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_6a92f0b3792a465a_v1", "elapsed_seconds": 21.486804962158203, "allocated": "6", "input_sha256": "5da1ee3cce399169645cbf03383f0c7594f9d9b878d0bc38e2e663bb1e06edd8", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 384, 2, 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": 3, "kernels": [ { "id": "0", "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": 8.96, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 46.570666, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 22.502958, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 50.005082, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 75.221106, "unit": "%" }, "dram__bytes_read.sum": { "value": 9.59744, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.043264, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.075971, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 18.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 1.024, "unit": "Kbyte/block" } } }, { "id": "1", "name": "void cutlass::Kernel2(Params)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 20.0, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 35.772448, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 12.288959, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 42.509332, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 16.049719, "unit": "%" }, "dram__bytes_read.sum": { "value": 9.906432, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 1.899264, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.590285, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 136.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 74.752, "unit": "Kbyte/block" } } }, { "id": "2", "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": 14.976, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 53.944847, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 27.158949, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 57.739165, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 76.844231, "unit": "%" }, "dram__bytes_read.sum": { "value": 19.185408, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.368896, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 1.305709, "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": "31b28e8acac4d172e8229285efd6b166430ed8b99750bb4d697e38d1e768012a", "metrics.csv": "127bc53381dd110a9af457e5b5bd55ea04be1f7d55642390aa1398d86f4b1b29", "replay_comparison.output.pt": "0a33b41fb849b3c194f0a6780abc4e6d0c3108b119250e5caf550581e9317f0c", "operator.ncu-rep": "7f787943d0c4cf69c0d5e78cbbd5a730aa6613fcaf7381a2ee87ab8c2106050e", "run.log": "71fdc8b3969ec3f203e23a920f008e4c2b217f4653c4d4de3e3484cda3f9ec6a", "identity.json": "defc0f33c11d51e29075f8ebaae3db347ed2613a098d86b5acd516ba2cf503c0", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "52d634e5cb939c801a29db0242e2eea8143c2c80c6c8a48bc0f1cc3d6883045a", "guard_own_pid_history.json": "e1ac78aafe774e9d10a4a4b1404de25a65f833247bc5fe82ba4f98104550865e" } }, { "stage": "vae", "status": "passed", "output": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/raw/ncu_missing_ddb198f9c3730e1d_v1", "elapsed_seconds": 19.37952160835266, "allocated": "6", "input_sha256": "480f421d034402c9b8b81d025dce79d2061ec150ba4509115415f8dd5d3b37b5", "correctness": { "op": "aten.conv3d.default", "stage": "vae", "shape": [ 1, 3, 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": 70.304, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 37.491615, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 63.85219, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 80.79348, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 91.921425, "unit": "%" }, "dram__bytes_read.sum": { "value": 116.620288, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 99.355648, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 3.072029, "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": 3.2, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.368443, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 0.266028, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 1.098645, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 11.62121, "unit": "%" }, "dram__bytes_read.sum": { "value": 0.040448, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.01264, "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": "sm80_xmma_fprop_implicit_gemm_indexed_wo_smem_tf32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x16x64_stage1_warpsize4x1x1_g1_tensor16x8x8_alignc4_execute_kernel__5x_cudnn", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 200.416, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 32.875367, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 12.640777, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 41.635844, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 24.081679, "unit": "%" }, "dram__bytes_read.sum": { "value": 117.859072, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 4.027904, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.60817, "unit": "Tbyte/s" }, "launch__registers_per_thread": { "value": 94.0, "unit": "register/thread" }, "launch__shared_mem_per_block": { "value": 5.12, "unit": "Kbyte/block" } } }, { "id": "3", "name": "void cudnn::nhwcToNchwKernel(cudnn::nhwc2nchw_params_t, const T1 *, T2 *)", "device": "0", "metrics": { "gpu__time_duration.sum": { "value": 6.496, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 53.903145, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 3.897877, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 11.51182, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 81.875792, "unit": "%" }, "dram__bytes_read.sum": { "value": 1.21344, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.186798, "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": 3.584, "unit": "us" }, "sm__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 15.263738, "unit": "%" }, "gpu__dram_throughput.avg.pct_of_peak_sustained_elapsed": { "value": 7.117743, "unit": "%" }, "lts__throughput.avg.pct_of_peak_sustained_elapsed": { "value": 16.514775, "unit": "%" }, "sm__warps_active.avg.pct_of_peak_sustained_active": { "value": 50.054617, "unit": "%" }, "dram__bytes_read.sum": { "value": 1.212672, "unit": "Mbyte" }, "dram__bytes_write.sum": { "value": 0.0, "unit": "Mbyte" }, "dram__bytes.sum.per_second": { "value": 0.338357, "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": "274c375dc3a4b0374b079502da113ad35e47b63ebdd6db9912b9260f0916f3b1", "metrics.csv": "01f2062c537480f807ac36f5686f41e50a309a38501119f779274fc8d8958fef", "replay_comparison.output.pt": "4a77bec5dde18ebe68e18379ddb0cb3ce8045c8f97f7b385750061356003d5ce", "operator.ncu-rep": "0e26dc4a20e669873fe3b9e90ce061119bc3bf64d897d4ba998d001c9a112668", "run.log": "faf4ed59b5b2c6f16ba027f637607a1605598c57b550f723d01b970ed7a8bff0", "identity.json": "d9cf7d5cc242b136aa7536b53ac9dc0f00bf5b098b4ed21791ecb693b817b327", "replay_operator.py": "cbea82d5e280ab2be0b1ce8dfebce6c1e80888c1d9fe727638e97a2652578d9b", "manifest.json": "00ed390ebd49a2fb4a95a5aeb191184e5d3a6d8d7e145685c24958f87162aa34", "guard_own_pid_history.json": "cb2fd5c5d65cb55ced98cf2f4cafd2c1be185f63f7504dd429b8e8ae148db7d5" } } ], "scope": "Thirteen additional 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." }