{
 "meta": {
  "date": "2026-10-22",
  "torch": "2.14.1+cu130",
  "driver": "595.91.07",
  "host": "RunPod pods (ARENA fleet)",
  "method": "Square matmuls n=8192 and n=16384 (best kept), ~2 s timed after warmup; FP8 via torch._scaled_mm (e4m3, bf16 out); INT8 via torch._int_mm with column-major B; DRAM = 1 GiB device copy (read+write counted). SM clock and power are medians of nvidia-smi samples during each test.",
  "script": "gpubench.py"
 },
 "cards": [
  {
   "id": "rtx-2000-ada",
   "power_limit_w": 70,
   "best": {
    "fp32": {
     "value": 6.0,
     "unit": "TFLOPS",
     "sm_mhz": 1710.0,
     "power_w": 69.3
    },
    "tf32": {
     "value": 17.14,
     "unit": "TFLOPS",
     "sm_mhz": 1635.0,
     "power_w": 69.4
    },
    "fp16": {
     "value": 35.46,
     "unit": "TFLOPS",
     "sm_mhz": 1665.0,
     "power_w": 69.4
    },
    "fp16_acc16": {
     "value": 38.49,
     "unit": "TFLOPS",
     "sm_mhz": 1755.0,
     "power_w": 69.3
    },
    "bf16": {
     "value": 36.4,
     "unit": "TFLOPS",
     "sm_mhz": 1710.0,
     "power_w": 69.6
    },
    "fp8": {
     "value": 73.76,
     "unit": "TFLOPS",
     "sm_mhz": 1740.0,
     "power_w": 69.5
    },
    "fp8_fastacc": {
     "value": 79.21,
     "unit": "TFLOPS",
     "sm_mhz": 1800.0,
     "power_w": 68.8
    },
    "int8": {
     "value": 83.85,
     "unit": "TOPS",
     "sm_mhz": 1980.0,
     "power_w": 68.4
    },
    "fp64": {
     "value": 0.19,
     "unit": "TFLOPS",
     "sm_mhz": 2415.0,
     "power_w": 28.4
    },
    "dram": {
     "value": 206.57,
     "unit": "GB/s",
     "sm_mhz": 2415.0,
     "power_w": 45.6
    }
   }
  },
  {
   "id": "rtx-a4000",
   "power_limit_w": 140,
   "best": {
    "fp32": {
     "value": 12.44,
     "unit": "TFLOPS",
     "sm_mhz": 1282.0,
     "power_w": 109.2
    },
    "tf32": {
     "value": 31.35,
     "unit": "TFLOPS",
     "sm_mhz": 1320.0,
     "power_w": 139.8
    },
    "fp16": {
     "value": 61.81,
     "unit": "TFLOPS",
     "sm_mhz": 1290.0,
     "power_w": 130.8
    },
    "fp16_acc16": {
     "value": 62.91,
     "unit": "TFLOPS",
     "sm_mhz": 1320.0,
     "power_w": 139.8
    },
    "bf16": {
     "value": 62.33,
     "unit": "TFLOPS",
     "sm_mhz": 1275.0,
     "power_w": 139.8
    },
    "int8": {
     "value": 116.94,
     "unit": "TOPS",
     "sm_mhz": 1268.0,
     "power_w": 139.7
    },
    "fp64": {
     "value": 0.32,
     "unit": "TFLOPS",
     "sm_mhz": 1950.0,
     "power_w": 102.6
    },
    "dram": {
     "value": 390.97,
     "unit": "GB/s",
     "sm_mhz": 1845.0,
     "power_w": 128.5
    }
   }
  },
  {
   "id": "rtx-a5000",
   "power_limit_w": 230.0,
   "best": {
    "fp32": {
     "value": 17.32,
     "unit": "TFLOPS",
     "sm_mhz": 1327.5,
     "power_w": 154.76999999999998
    },
    "tf32": {
     "value": 44.57,
     "unit": "TFLOPS",
     "sm_mhz": 1395.0,
     "power_w": 227.47
    },
    "fp16": {
     "value": 88.33,
     "unit": "TFLOPS",
     "sm_mhz": 1380.0,
     "power_w": 213.13
    },
    "fp16_acc16": {
     "value": 88.76,
     "unit": "TFLOPS",
     "sm_mhz": 1395.0,
     "power_w": 229.2
    },
    "bf16": {
     "value": 90.62,
     "unit": "TFLOPS",
     "sm_mhz": 1417.5,
     "power_w": 222.46
    },
    "int8": {
     "value": 166.62,
     "unit": "TOPS",
     "sm_mhz": 1365.0,
     "power_w": 228.64
    },
    "fp64": {
     "value": 0.42,
     "unit": "TFLOPS",
     "sm_mhz": 1890.0,
     "power_w": 137.51
    },
    "dram": {
     "value": 683.98,
     "unit": "GB/s",
     "sm_mhz": 1785.0,
     "power_w": 198.45999999999998
    }
   }
  },
  {
   "id": "a40",
   "power_limit_w": 300.0,
   "best": {
    "fp32": {
     "value": 23.31,
     "unit": "TFLOPS",
     "sm_mhz": 1395.0,
     "power_w": 297.15999999999997
    },
    "tf32": {
     "value": 58.19,
     "unit": "TFLOPS",
     "sm_mhz": 1440.0,
     "power_w": 299.71000000000004
    },
    "fp16": {
     "value": 118.11,
     "unit": "TFLOPS",
     "sm_mhz": 1380.0,
     "power_w": 299.2
    },
    "fp16_acc16": {
     "value": 117.85,
     "unit": "TFLOPS",
     "sm_mhz": 1410.0,
     "power_w": 300.44
    },
    "bf16": {
     "value": 122.04,
     "unit": "TFLOPS",
     "sm_mhz": 1440.0,
     "power_w": 299.2
    },
    "int8": {
     "value": 227.06,
     "unit": "TOPS",
     "sm_mhz": 1395.0,
     "power_w": 300.15
    },
    "fp64": {
     "value": 0.5,
     "unit": "TFLOPS",
     "sm_mhz": 1740.0,
     "power_w": 134.24
    },
    "dram": {
     "value": 562.97,
     "unit": "GB/s",
     "sm_mhz": 1740.0,
     "power_w": 212.93
    }
   }
  }
 ]
}