{ "scenario": "RVA", "backend": "cudnn", "sequence": 81547, "batch": 1, "heads": 28, "head_dim": 128, "pid": 2815990, "torch": "2.11.0+cu130", "cuda": "13.0", "cudnn": 91900, "gpu": "NVIDIA RTX 6000D", "capability": [ 12, 0 ], "started_at": 1787624713.1974936, "results": [ { "layout": "model_strided", "dtype": "torch.bfloat16", "q_shape": [ 1, 28, 81547, 128 ], "q_stride": [ 876793344, 128, 10752, 1 ], "k_stride": [ 876793344, 128, 10752, 1 ], "v_stride": [ 876793344, 128, 10752, 1 ], "q_contiguous": false, "warm": { "samples_ms": [ 663.4467163085938, 663.6146850585938, 663.8494873046875, 665.0116577148438, 666.3472290039062, 666.7335205078125, 666.774169921875, 666.885009765625, 666.9378051757812, 667.082763671875, 667.2296142578125, 667.2323608398438 ], "min_ms": 663.4467163085938, "median_ms": 666.7538452148438, "mean_ms": 665.9287516276041, "p90_ms": 667.2296142578125, "max_ms": 667.2323608398438, "stdev_ms": 1.5037322109933906, "cv": 0.0022580977429163427 }, "cold": { "samples_ms": [ 661.3488159179688, 662.2349243164062, 662.9520874023438, 663.33935546875, 664.3153076171875 ], "min_ms": 661.3488159179688, "median_ms": 662.9520874023438, "mean_ms": 662.8380981445313, "p90_ms": 664.3153076171875, "max_ms": 664.3153076171875, "stdev_ms": 1.1211802374478943, "cv": 0.0016914843015004577 }, "cold_over_warm": 0.9942981089051313, "effective_tflops": 142.98103632759015, "io_lower_bound_gbps": 3.5067148105409194, "cuda_graph": { "supported": true, "samples_ms": [ 660.9876708984375, 661.070068359375, 661.7693481445312, 661.9100952148438, 662.0004272460938, 662.1380004882812, 662.2643432617188, 662.8046264648438, 663.14208984375, 663.273193359375, 663.3394165039062, 664.5430297851562 ], "min_ms": 660.9876708984375, "median_ms": 662.201171875, "mean_ms": 662.4368591308594, "p90_ms": 663.3394165039062, "max_ms": 664.5430297851562, "stdev_ms": 1.0261795181545272, "cv": 0.0015490978559087295 }, "output_checksum": 8.008569238882046e-07, "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu5-RVA-cudnn-model_strided.trace.json", "trace_error": null }, { "layout": "contiguous", "dtype": "torch.bfloat16", "q_shape": [ 1, 28, 81547, 128 ], "q_stride": [ 292264448, 10438016, 128, 1 ], "k_stride": [ 292264448, 10438016, 128, 1 ], "v_stride": [ 292264448, 10438016, 128, 1 ], "q_contiguous": true, "warm": { "samples_ms": [ 660.9556274414062, 661.1549072265625, 661.227783203125, 661.380615234375, 662.6010131835938, 663.0504150390625, 663.0765991210938, 663.2205200195312, 663.263427734375, 663.3666381835938, 663.4912719726562, 664.0556640625 ], "min_ms": 660.9556274414062, "median_ms": 663.0635070800781, "mean_ms": 662.5703735351562, "p90_ms": 663.4912719726562, "max_ms": 664.0556640625, "stdev_ms": 1.0832552060899807, "cv": 0.0016349285288900753 }, "cold": { "samples_ms": [ 661.0634765625, 662.4185180664062, 662.55810546875, 663.3753662109375, 663.77783203125 ], "min_ms": 661.0634765625, "median_ms": 662.55810546875, "mean_ms": 662.6386596679688, "p90_ms": 663.77783203125, "max_ms": 663.77783203125, "stdev_ms": 1.0460166493920169, "cv": 0.0015785626662895716 }, "cold_over_warm": 0.9992377779715946, "effective_tflops": 143.77680983235084, "io_lower_bound_gbps": 3.5262317395453135, "cuda_graph": { "supported": true, "samples_ms": [ 663.2074584960938, 663.2119750976562, 663.25390625, 663.2601318359375, 663.2670288085938, 663.302978515625, 663.3111572265625, 663.311279296875, 663.3232421875, 663.3523559570312, 663.3528442382812, 663.3601684570312 ], "min_ms": 663.2074584960938, "median_ms": 663.3070678710938, "mean_ms": 663.2928771972656, "p90_ms": 663.3528442382812, "max_ms": 663.3601684570312, "stdev_ms": 0.05262162134716431, "cv": 7.933391591587143e-05 }, "output_checksum": 5.330828685146116e-07, "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu5-RVA-cudnn-contiguous.trace.json", "trace_error": null } ], "status": "success", "finished_at": 1787624759.6317039 }