h3-blackwell-runtime/benchmarks/gb10-post-fc2-block24-targeted-summary-20260826.json

{
  "source_csv": "benchmarks\\gb10-post-fc2-block24-targeted-20260826.csv",
  "source_traffic_csv": "benchmarks\\gb10-post-fc2-block24-targeted-traffic-20260826.csv",
  "launch_order_contract": [
    "qkv",
    "sage2",
    "attention_output",
    "fc1",
    "fc2"
  ],
  "cache_control": "none (warmed/uncontrolled cache, as reported by NCU)",
  "metrics": {
    "qkv": {
      "launch_id": 0,
      "kernel_name": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu",
      "grid_size": "(296, 168, 1)",
      "block_size": "(384, 1, 1)",
      "duration_ns": 26376032.0,
      "registers_per_thread": 168.0,
      "achieved_occupancy_percent": 20.83,
      "eligible_warps_per_scheduler": 0.24,
      "issue_active_percent": 18.29,
      "sm_throughput_percent": 78.99,
      "tensor_pipe_active_percent": 78.99,
      "l2_requested_bytes": 40128950176.0,
      "l2_hit_rate_percent": 89.11,
      "l2_throughput_percent": 77.1,
      "memory_throughput_percent": 76.92,
      "local_spilling_requests": 0.0,
      "top_scheduler_stalls": [
        {
          "reason": "sleeping",
          "warps_per_issue_active": 3.93
        },
        {
          "reason": "math_pipe_throttle",
          "warps_per_issue_active": 3.82
        },
        {
          "reason": "wait",
          "warps_per_issue_active": 3.78
        },
        {
          "reason": "selected",
          "warps_per_issue_active": 1.0
        },
        {
          "reason": "long_scoreboard",
          "warps_per_issue_active": 0.61
        }
      ]
    },
    "sage2": {
      "launch_id": 1,
      "kernel_name": "void qk_int_sv_f8_attn_kernel<128, 64, 32, 64, 128, 1, 2, 2, float, 1, __nv_bfloat16, 1, 0, 0, 1, 0, 1>(signed char *, signed char *, signed char *, T11 *, float *, float *, float *, float *, float *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, float)",
      "grid_size": "(296, 56, 1)",
      "block_size": "(32, 4, 1)",
      "duration_ns": 258996000.0,
      "registers_per_thread": 255.0,
      "achieved_occupancy_percent": 16.65,
      "eligible_warps_per_scheduler": 0.46,
      "issue_active_percent": 36.49,
      "sm_throughput_percent": 75.51,
      "tensor_pipe_active_percent": 75.51,
      "l2_requested_bytes": 161497982464.0,
      "l2_hit_rate_percent": 98.85,
      "l2_throughput_percent": 31.59,
      "memory_throughput_percent": 31.74,
      "local_spilling_requests": 1458688.0,
      "top_scheduler_stalls": [
        {
          "reason": "wait",
          "warps_per_issue_active": 2.01
        },
        {
          "reason": "math_pipe_throttle",
          "warps_per_issue_active": 1.24
        },
        {
          "reason": "selected",
          "warps_per_issue_active": 1.0
        },
        {
          "reason": "not_selected",
          "warps_per_issue_active": 0.26
        },
        {
          "reason": "short_scoreboard",
          "warps_per_issue_active": 0.25
        }
      ]
    },
    "attention_output": {
      "launch_id": 2,
      "kernel_name": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu",
      "grid_size": "(296, 42, 1)",
      "block_size": "(384, 1, 1)",
      "duration_ns": 8481152.0,
      "registers_per_thread": 168.0,
      "achieved_occupancy_percent": 20.83,
      "eligible_warps_per_scheduler": 0.23,
      "issue_active_percent": 17.51,
      "sm_throughput_percent": 81.88,
      "tensor_pipe_active_percent": 81.88,
      "l2_requested_bytes": 13241354176.0,
      "l2_hit_rate_percent": 89.89,
      "l2_throughput_percent": 79.1,
      "memory_throughput_percent": 79.51,
      "local_spilling_requests": 0.0,
      "top_scheduler_stalls": [
        {
          "reason": "sleeping",
          "warps_per_issue_active": 4.08
        },
        {
          "reason": "math_pipe_throttle",
          "warps_per_issue_active": 3.9
        },
        {
          "reason": "wait",
          "warps_per_issue_active": 3.85
        },
        {
          "reason": "selected",
          "warps_per_issue_active": 1.0
        },
        {
          "reason": "long_scoreboard",
          "warps_per_issue_active": 0.6
        }
      ]
    },
    "fc1": {
      "launch_id": 3,
      "kernel_name": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu",
      "grid_size": "(296, 224, 1)",
      "block_size": "(384, 1, 1)",
      "duration_ns": 36172608.0,
      "registers_per_thread": 168.0,
      "achieved_occupancy_percent": 20.83,
      "eligible_warps_per_scheduler": 0.23,
      "issue_active_percent": 17.39,
      "sm_throughput_percent": 76.85,
      "tensor_pipe_active_percent": 76.85,
      "l2_requested_bytes": 53505320928.0,
      "l2_hit_rate_percent": 89.11,
      "l2_throughput_percent": 74.99,
      "memory_throughput_percent": 74.85,
      "local_spilling_requests": 0.0,
      "top_scheduler_stalls": [
        {
          "reason": "sleeping",
          "warps_per_issue_active": 4.41
        },
        {
          "reason": "math_pipe_throttle",
          "warps_per_issue_active": 3.82
        },
        {
          "reason": "wait",
          "warps_per_issue_active": 3.78
        },
        {
          "reason": "selected",
          "warps_per_issue_active": 1.0
        },
        {
          "reason": "long_scoreboard",
          "warps_per_issue_active": 0.61
        }
      ]
    },
    "fc2": {
      "launch_id": 4,
      "kernel_name": "cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu",
      "grid_size": "(296, 42, 1)",
      "block_size": "(384, 1, 1)",
      "duration_ns": 16198976.0,
      "registers_per_thread": 168.0,
      "achieved_occupancy_percent": 20.83,
      "eligible_warps_per_scheduler": 0.23,
      "issue_active_percent": 17.44,
      "sm_throughput_percent": 85.88,
      "tensor_pipe_active_percent": 85.88,
      "l2_requested_bytes": 26072923840.0,
      "l2_hit_rate_percent": 91.06,
      "l2_throughput_percent": 81.58,
      "memory_throughput_percent": 77.65,
      "local_spilling_requests": 0.0,
      "top_scheduler_stalls": [
        {
          "reason": "math_pipe_throttle",
          "warps_per_issue_active": 4.06
        },
        {
          "reason": "wait",
          "warps_per_issue_active": 3.98
        },
        {
          "reason": "sleeping",
          "warps_per_issue_active": 3.81
        },
        {
          "reason": "selected",
          "warps_per_issue_active": 1.0
        },
        {
          "reason": "long_scoreboard",
          "warps_per_issue_active": 0.6
        }
      ]
    }
  },
  "units": {
    "duration_ns": "ns",
    "l2_requested_bytes": "lts__t_bytes.sum",
    "throughput_and_hit_rate": "%"
  }
}