{ "scenario": "RVA", "backend": "flash", "sequence": 81547, "batch": 1, "heads": 28, "head_dim": 128, "pid": 2815989, "torch": "2.11.0+cu130", "cuda": "13.0", "cudnn": 91900, "gpu": "NVIDIA RTX 6000D", "capability": [ 12, 0 ], "started_at": 1787624713.0858145, "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": [ 668.9849243164062, 669.0095825195312, 669.1596069335938, 669.2590942382812, 669.2791137695312, 669.3319091796875, 669.3397216796875, 669.34619140625, 669.434326171875, 669.5739135742188, 669.793701171875, 669.866455078125 ], "min_ms": 668.9849243164062, "median_ms": 669.3358154296875, "mean_ms": 669.3648783365885, "p90_ms": 669.793701171875, "max_ms": 669.866455078125, "stdev_ms": 0.2725866615400381, "cv": 0.0004072317959338293 }, "cold": { "samples_ms": [ 666.1293334960938, 667.8314208984375, 668.2146606445312, 668.2601318359375, 668.8970336914062 ], "min_ms": 666.1293334960938, "median_ms": 668.2146606445312, "mean_ms": 667.8665161132812, "p90_ms": 668.8970336914062, "max_ms": 668.8970336914062, "stdev_ms": 1.0436881179439383, "cv": 0.0015627196344829353 }, "cold_over_warm": 0.998324974161981, "effective_tflops": 142.42948541909988, "io_lower_bound_gbps": 3.4931876198781038, "cuda_graph": { "supported": true, "samples_ms": [ 665.1121826171875, 665.2501831054688, 665.33447265625, 665.3963623046875, 665.561767578125, 665.6658935546875, 665.684814453125, 665.8719482421875, 666.2486572265625, 666.40673828125, 666.5081176757812, 666.615478515625 ], "min_ms": 665.1121826171875, "median_ms": 665.6753540039062, "mean_ms": 665.8047180175781, "p90_ms": 666.5081176757812, "max_ms": 666.615478515625, "stdev_ms": 0.5208216704522183, "cv": 0.0007822438867705169 }, "output_checksum": 8.005773111108283e-07, "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu4-RVA-flash-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": [ 665.3004760742188, 665.3120727539062, 665.370849609375, 665.6288452148438, 665.7316284179688, 665.8421630859375, 665.8939208984375, 666.1165771484375, 666.1333618164062, 666.437255859375, 666.5823974609375, 666.8427124023438 ], "min_ms": 665.3004760742188, "median_ms": 665.8680419921875, "mean_ms": 665.9326883951823, "p90_ms": 666.5823974609375, "max_ms": 666.8427124023438, "stdev_ms": 0.5064943096262162, "cv": 0.000760578836949432 }, "cold": { "samples_ms": [ 665.0519409179688, 665.1027221679688, 665.4036254882812, 665.5978393554688, 665.7393188476562 ], "min_ms": 665.0519409179688, "median_ms": 665.4036254882812, "mean_ms": 665.3790893554688, "p90_ms": 665.7393188476562, "max_ms": 665.7393188476562, "stdev_ms": 0.3006758195467372, "cv": 0.00045188648750292434 }, "cold_over_warm": 0.9993025397306697, "effective_tflops": 143.1712437782718, "io_lower_bound_gbps": 3.5113797878100184, "cuda_graph": { "supported": true, "samples_ms": [ 665.2788696289062, 665.302001953125, 665.3150024414062, 665.7102661132812, 665.90283203125, 665.9227294921875, 665.944091796875, 665.9635009765625, 666.3156127929688, 666.32421875, 666.3789672851562, 666.448486328125 ], "min_ms": 665.2788696289062, "median_ms": 665.9334106445312, "mean_ms": 665.9005482991537, "p90_ms": 666.3789672851562, "max_ms": 666.448486328125, "stdev_ms": 0.4275794078744464, "cv": 0.0006421070067693618 }, "output_checksum": 5.333090484782588e-07, "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu4-RVA-flash-contiguous.trace.json", "trace_error": null } ], "status": "success", "finished_at": 1787624759.749213 }