{
  "harness": "benchmarks/sr_forward_pass/bench_forward_pass.py",
  "measures": "bare SPAN forward pass vs per-stage pipeline cost, batch 1",
  "scale_factor": 2,
  "weights_path": "/models/span/2xNomosUni_span_multijpg.safetensors",
  "feature_channels": 48,
  "half_precision": true,
  "device": "cuda",
  "torch_version": "2.14.0.dev20260617+cu130",
  "warmup_iters": 20,
  "timed_iters": 100,
  "canvas": "1920x1080",
  "box_state_at_start": {
    "captured_at": 1785921697.0823193,
    "load_average": [
      63.37,
      62.86,
      61.96
    ],
    "cpu_count": 20,
    "mem_total_mib": 124611,
    "mem_available_mib": 51139,
    "gpu_name": "NVIDIA GB10",
    "note": "Co-tenant load on the host at this instant. Forward-pass cost can only be compared across runs taken under comparable box state."
  },
  "resolutions": [
    {
      "input_w": 640,
      "input_h": 360,
      "model_forward": {
        "n": 100,
        "mean_ms": 447.329,
        "p50_ms": 446.403,
        "p95_ms": 457.215,
        "p99_ms": 464.38,
        "min_ms": 433.937,
        "max_ms": 469.559,
        "implied_fps_at_mean": 2.24
      },
      "upscale_call": {
        "n": 100,
        "mean_ms": 411.696,
        "p50_ms": 433.964,
        "p95_ms": 494.882,
        "p99_ms": 517.912,
        "min_ms": 327.452,
        "max_ms": 527.944,
        "implied_fps_at_mean": 2.43
      },
      "sr_output_shape": "1280x720",
      "fit_to_canvas": {
        "n": 100,
        "mean_ms": 75.557,
        "p50_ms": 59.603,
        "p95_ms": 160.049,
        "p99_ms": 267.134,
        "min_ms": 9.148,
        "max_ms": 367.858,
        "implied_fps_at_mean": 13.23
      },
      "sharpness_probe": {
        "n": 100,
        "mean_ms": 156.794,
        "p50_ms": 119.501,
        "p95_ms": 316.125,
        "p99_ms": 411.951,
        "min_ms": 71.815,
        "max_ms": 435.919,
        "implied_fps_at_mean": 6.38
      },
      "attributed": {
        "per_frame_ms": 644.047,
        "implied_fps": 1.55,
        "model_share_pct": 69.5,
        "overhead_share_pct": 30.5
      }
    },
    {
      "input_w": 854,
      "input_h": 480,
      "model_forward": {
        "n": 100,
        "mean_ms": 575.281,
        "p50_ms": 560.089,
        "p95_ms": 621.347,
        "p99_ms": 792.046,
        "min_ms": 540.244,
        "max_ms": 817.151,
        "implied_fps_at_mean": 1.74
      },
      "upscale_call": {
        "n": 100,
        "mean_ms": 813.119,
        "p50_ms": 637.572,
        "p95_ms": 1190.159,
        "p99_ms": 1455.12,
        "min_ms": 586.281,
        "max_ms": 2534.612,
        "implied_fps_at_mean": 1.23
      },
      "sr_output_shape": "1708x960",
      "fit_to_canvas": {
        "n": 100,
        "mean_ms": 59.926,
        "p50_ms": 54.182,
        "p95_ms": 103.212,
        "p99_ms": 145.383,
        "min_ms": 16.247,
        "max_ms": 228.179,
        "implied_fps_at_mean": 16.69
      },
      "sharpness_probe": {
        "n": 100,
        "mean_ms": 101.225,
        "p50_ms": 86.57,
        "p95_ms": 195.172,
        "p99_ms": 243.422,
        "min_ms": 40.917,
        "max_ms": 313.348,
        "implied_fps_at_mean": 9.88
      },
      "attributed": {
        "per_frame_ms": 974.27,
        "implied_fps": 1.03,
        "model_share_pct": 59.0,
        "overhead_share_pct": 41.0
      }
    },
    {
      "input_w": 1280,
      "input_h": 720,
      "model_forward": {
        "n": 100,
        "mean_ms": 1234.137,
        "p50_ms": 1228.804,
        "p95_ms": 1282.784,
        "p99_ms": 1328.554,
        "min_ms": 1207.93,
        "max_ms": 1350.739,
        "implied_fps_at_mean": 0.81
      },
      "upscale_call": {
        "n": 100,
        "mean_ms": 1830.958,
        "p50_ms": 1698.014,
        "p95_ms": 2605.395,
        "p99_ms": 3030.503,
        "min_ms": 1315.322,
        "max_ms": 3803.661,
        "implied_fps_at_mean": 0.55
      },
      "sr_output_shape": "2560x1440",
      "fit_to_canvas": {
        "n": 100,
        "mean_ms": 35.401,
        "p50_ms": 31.319,
        "p95_ms": 63.027,
        "p99_ms": 80.392,
        "min_ms": 11.385,
        "max_ms": 84.426,
        "implied_fps_at_mean": 28.25
      },
      "sharpness_probe": {
        "n": 100,
        "mean_ms": 107.447,
        "p50_ms": 81.838,
        "p95_ms": 253.873,
        "p99_ms": 328.376,
        "min_ms": 50.776,
        "max_ms": 382.115,
        "implied_fps_at_mean": 9.31
      },
      "attributed": {
        "per_frame_ms": 1973.806,
        "implied_fps": 0.51,
        "model_share_pct": 62.5,
        "overhead_share_pct": 37.5
      }
    }
  ],
  "caveat": "SPAN's published 100+ fps is the authors' capability figure on their hardware. It is not a GB10 pipeline result and is not a target this project set or missed.",
  "box_state_at_end": {
    "captured_at": 1785922386.1958196,
    "load_average": [
      59.47,
      69.52,
      67.58
    ],
    "cpu_count": 20,
    "mem_total_mib": 124611,
    "mem_available_mib": 52793,
    "gpu_name": "NVIDIA GB10",
    "note": "Co-tenant load on the host at this instant. Forward-pass cost can only be compared across runs taken under comparable box state."
  },
  "duration_s": 694.92,
  "arm": "as-deployed"
}