outputs/nvfp4/benchmark.json

{
  "bf16": [
    {
      "batch": 1,
      "res": "512x512",
      "total_s": 0.467,
      "img_per_s": 2.142,
      "step_ms": 92.2,
      "vram_gb": 16.6
    },
    {
      "batch": 2,
      "res": "512x512",
      "total_s": 0.952,
      "img_per_s": 2.101,
      "step_ms": 191.9,
      "vram_gb": 17.1
    },
    {
      "batch": 4,
      "res": "512x512",
      "total_s": 2.014,
      "img_per_s": 1.986,
      "step_ms": 411.2,
      "vram_gb": 18.2
    },
    {
      "batch": 1,
      "res": "1024x1024",
      "total_s": 1.589,
      "img_per_s": 0.629,
      "step_ms": 332.0,
      "vram_gb": 18.6
    },
    {
      "batch": 2,
      "res": "1024x1024",
      "total_s": 3.402,
      "img_per_s": 0.588,
      "step_ms": 706.1,
      "vram_gb": 20.3
    },
    {
      "batch": 4,
      "res": "1024x1024",
      "total_s": 6.799,
      "img_per_s": 0.588,
      "step_ms": 1417.7,
      "vram_gb": 24.6
    }
  ],
  "nvfp4_fused": [
    {
      "batch": 1,
      "res": "512x512",
      "total_s": 0.268,
      "img_per_s": 3.728,
      "step_ms": 42.4,
      "vram_gb": 12.6,
      "speedup_vs_bf16": 1.743,
      "vram_reduction": 0.241
    },
    {
      "batch": 1,
      "res": "1024x1024",
      "total_s": 0.835,
      "img_per_s": 1.198,
      "step_ms": 146.4,
      "vram_gb": 14.5,
      "speedup_vs_bf16": 1.903,
      "vram_reduction": 0.22
    }
  ]
}