{
 "date": "2026-10-21",
 "change": "Rejected: shared per-phase arrival word (one acquire instead of one per sender) with the FFN dispatch wait moved to an idle dispatch warp at kernel entry",
 "native_source_sha256": "see per-screen metadata; scratch tree HEAD 78a442f + patch",
 "decision": "not shipped: about 1% with repeats disagreeing in two of three layouts; 1A:3F T=1 slower in 3 of 5 repeats",
 "screens": {
  "sa-ffnranks2attentionranks2": {
   "arms": {
    "afd": {},
    "serial": {
     "_env": {
      "AFDEP_SERIAL_WAIT": "1"
     }
    },
    "masked": {
     "role_aware": false
    }
   },
   "passthrough": [
    "--payload",
    "bf16",
    "--routing",
    "random",
    "--hidden",
    "2048",
    "--ffn-ranks",
    "2",
    "--attention-ranks",
    "2",
    "--tokens",
    "1",
    "8",
    "32",
    "128",
    "512",
    "2048",
    "--capacity",
    "2048",
    "--warmup",
    "50",
    "--iterations",
    "200",
    "--blocks",
    "2"
   ],
   "processes": [
    {
     "order": [
      "afd",
      "serial",
      "masked"
     ],
     "median_us": {
      "afd": {
       "1": 20.735520124435425,
       "8": 23.741120100021362,
       "32": 25.697200298309326,
       "128": 35.616639852523804,
       "512": 70.67080020904541,
       "2048": 191.92415237426758
      },
      "serial": {
       "1": 20.77072024345398,
       "8": 24.63279962539673,
       "32": 25.797359943389893,
       "128": 35.71951985359192,
       "512": 70.52248001098633,
       "2048": 192.34383583068848
      },
      "masked": {
       "1": 31.666879653930664,
       "8": 34.131840467453,
       "32": 35.17975926399231,
       "128": 44.44551944732666,
       "512": 78.193039894104,
       "2048": 200.39687156677246
      }
     }
    },
    {
     "order": [
      "serial",
      "masked",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 21.096479892730713,
       "8": 24.663039445877075,
       "32": 26.11456036567688,
       "128": 36.313759088516235,
       "512": 70.63783884048462,
       "2048": 192.0453643798828
      },
      "masked": {
       "1": 31.45967960357666,
       "8": 35.43184041976929,
       "32": 35.46112060546875,
       "128": 44.21663999557495,
       "512": 78.36672067642212,
       "2048": 201.78600311279297
      },
      "afd": {
       "1": 20.999040603637695,
       "8": 24.627039432525635,
       "32": 26.150399446487427,
       "128": 35.537039041519165,
       "512": 70.04119873046875,
       "2048": 192.38207817077637
      }
     }
    },
    {
     "order": [
      "masked",
      "afd",
      "serial"
     ],
     "median_us": {
      "masked": {
       "1": 31.37368083000183,
       "8": 34.32103991508484,
       "32": 35.97583889961243,
       "128": 44.55079793930054,
       "512": 78.88848066329956,
       "2048": 200.4580020904541
      },
      "afd": {
       "1": 20.929759740829468,
       "8": 24.467040300369263,
       "32": 25.96160054206848,
       "128": 35.96735954284668,
       "512": 70.43896198272705,
       "2048": 192.4733543395996
      },
      "serial": {
       "1": 21.11016035079956,
       "8": 25.028001070022583,
       "32": 26.294400691986084,
       "128": 35.75999975204468,
       "512": 70.85911989212036,
       "2048": 192.45840072631836
      }
     }
    }
   ],
   "table": [
    {
     "tokens": 1,
     "afd": {
      "median_us": 20.929759740829468,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 21.096479892730713,
      "paired_change_pct": [
       0.1697575889455294,
       0.4640178136335438,
       0.8619334966285752
      ],
      "geomean_change_pct": 0.4981695224125726,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 31.45967960357666,
      "paired_change_pct": [
       52.718038727243524,
       49.81484248440784,
       49.89986133858248
      ],
      "geomean_change_pct": 50.80490603595946,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 8,
     "afd": {
      "median_us": 24.467040300369263,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 24.663039445877075,
      "paired_change_pct": [
       3.755844381472828,
       0.14618084098201312,
       2.2927201768856875
      ],
      "geomean_change_pct": 2.054124844000804,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 34.32103991508484,
      "paired_change_pct": [
       43.76676552603889,
       43.873730810587986,
       40.2745877463849
      ],
      "geomean_change_pct": 42.62850701803011,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 32,
     "afd": {
      "median_us": 25.96160054206848,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 26.11456036567688,
      "paired_change_pct": [
       0.38976870599851665,
       -0.13704984080218363,
       1.2818938084280695
      ],
      "geomean_change_pct": 0.5098333607858718,
      "repeats_faster": 1
     },
     "masked": {
      "median_us": 35.46112060546875,
      "paired_change_pct": [
       36.90113652695022,
       35.60450836720186,
       38.573270324057084
      ],
      "geomean_change_pct": 37.020921154944794,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 128,
     "afd": {
      "median_us": 35.616639852523804,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 35.75999975204468,
      "paired_change_pct": [
       0.28885375345373276,
       2.185663375301483,
       -0.5765221396221265
      ],
      "geomean_change_pct": 0.626074073415217,
      "repeats_faster": 1
     },
     "masked": {
      "median_us": 44.44551944732666,
      "paired_change_pct": [
       24.788637084689057,
       24.42409719030081,
       23.86452190416899
      ],
      "geomean_change_pct": 24.358504335953036,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 512,
     "afd": {
      "median_us": 70.43896198272705,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 70.63783884048462,
      "paired_change_pct": [
       -0.20987479640862672,
       0.8518416600947321,
       0.5964850951329259
      ],
      "geomean_change_pct": 0.41179615509419065,
      "repeats_faster": 1
     },
     "masked": {
      "median_us": 78.36672067642212,
      "paired_change_pct": [
       10.644056191252528,
       11.88660687831955,
       11.995518449923349
      ],
      "geomean_change_pct": 11.507037791746445,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 2048,
     "afd": {
      "median_us": 192.38207817077637,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 192.34383583068848,
      "paired_change_pct": [
       0.21867151748702351,
       -0.17502347105048388,
       -0.0077691861985584865
      ],
      "geomean_change_pct": 0.011829524699802363,
      "repeats_faster": 2
     },
     "masked": {
      "median_us": 200.4580020904541,
      "paired_change_pct": [
       4.414618529085601,
       4.888150201636132,
       4.148443184902573
      ],
      "geomean_change_pct": 4.483289753672404,
      "repeats_faster": 0
     }
    }
   ],
   "pod": "afdep-v21-h100-c",
   "python": "3.12.3 (main, Aug 31 2026, 10:18:26) [GCC 13.3.0]"
  },
  "sa-ffnranks1attentionranks3": {
   "arms": {
    "afd": {},
    "serial": {
     "_env": {
      "AFDEP_SERIAL_WAIT": "1"
     }
    },
    "masked": {
     "role_aware": false
    }
   },
   "passthrough": [
    "--payload",
    "bf16",
    "--routing",
    "random",
    "--hidden",
    "2048",
    "--ffn-ranks",
    "1",
    "--attention-ranks",
    "3",
    "--tokens",
    "1",
    "8",
    "32",
    "128",
    "512",
    "2048",
    "--capacity",
    "2048",
    "--warmup",
    "50",
    "--iterations",
    "200",
    "--blocks",
    "2"
   ],
   "processes": [
    {
     "order": [
      "afd",
      "serial",
      "masked"
     ],
     "median_us": {
      "afd": {
       "1": 24.61311936378479,
       "8": 26.954959630966187,
       "32": 32.813600301742554,
       "128": 45.28111934661865,
       "512": 108.97799968719482,
       "2048": 367.2813606262207
      },
      "serial": {
       "1": 24.310638904571533,
       "8": 27.134079933166504,
       "32": 33.309279680252075,
       "128": 46.58095836639404,
       "512": 109.37503814697266,
       "2048": 368.38951110839844
      },
      "masked": {
       "1": 36.56775951385498,
       "8": 38.119120597839355,
       "32": 44.43471908569336,
       "128": 55.41600227355957,
       "512": 118.52503776550293,
       "2048": 371.5630340576172
      }
     }
    },
    {
     "order": [
      "serial",
      "masked",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 25.493040084838867,
       "8": 28.176239728927612,
       "32": 34.717040061950684,
       "128": 46.514480113983154,
       "512": 109.44015979766846,
       "2048": 368.5086441040039
      },
      "masked": {
       "1": 35.72496056556702,
       "8": 38.334959745407104,
       "32": 43.68592023849487,
       "128": 55.19888162612915,
       "512": 117.84783840179443,
       "2048": 370.4514503479004
      },
      "afd": {
       "1": 24.72007989883423,
       "8": 27.61415958404541,
       "32": 33.14695954322815,
       "128": 45.45352220535278,
       "512": 108.92752170562744,
       "2048": 366.67598724365234
      }
     }
    },
    {
     "order": [
      "masked",
      "afd",
      "serial"
     ],
     "median_us": {
      "masked": {
       "1": 36.22200012207031,
       "8": 39.85551953315735,
       "32": 44.74463939666748,
       "128": 55.715439319610596,
       "512": 119.5308780670166,
       "2048": 370.8937644958496
      },
      "afd": {
       "1": 24.708479642868042,
       "8": 27.234959602355957,
       "32": 33.044880628585815,
       "128": 45.46895980834961,
       "512": 109.54848289489746,
       "2048": 366.6192054748535
      },
      "serial": {
       "1": 25.030640363693237,
       "8": 28.051199913024902,
       "32": 34.538559913635254,
       "128": 46.34479999542236,
       "512": 109.50600147247314,
       "2048": 368.4142303466797
      }
     }
    }
   ],
   "table": [
    {
     "tokens": 1,
     "afd": {
      "median_us": 24.708479642868042,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 25.030640363693237,
      "paired_change_pct": [
       -1.2289399597936401,
       3.126851487405946,
       1.3038467986766111
      ],
      "geomean_change_pct": 1.051431514399126,
      "repeats_faster": 1
     },
     "masked": {
      "median_us": 36.22200012207031,
      "paired_change_pct": [
       48.57019532298692,
       44.51798178553528,
       46.59744608173648
      ],
      "geomean_change_pct": 46.55253291393204,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 8,
     "afd": {
      "median_us": 27.234959602355957,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 28.051199913024902,
      "paired_change_pct": [
       0.6645170486345053,
       2.0354780060261257,
       2.997031471999456
      ],
      "geomean_change_pct": 1.8945076538355865,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 38.334959745407104,
      "paired_change_pct": [
       41.41783597422881,
       38.823561255711134,
       46.33955810865109
      ],
      "geomean_change_pct": 42.15965775391397,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 32,
     "afd": {
      "median_us": 33.044880628585815,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 34.538559913635254,
      "paired_change_pct": [
       1.5105912607925598,
       4.736725601257441,
       4.520153369104074
      ],
      "geomean_change_pct": 3.578622107449725,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 44.43471908569336,
      "paired_change_pct": [
       35.41555537059939,
       31.79465278413389,
       35.40566207390281
      ],
      "geomean_change_pct": 34.19439916301741,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 128,
     "afd": {
      "median_us": 45.45352220535278,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 46.514480113983154,
      "paired_change_pct": [
       2.8705982505100147,
       2.3341599443869487,
       1.9262375712230773
      ],
      "geomean_change_pct": 2.376268479611676,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 55.41600227355957,
      "paired_change_pct": [
       22.382138677624663,
       21.440273378041354,
       22.535108686122605
      ],
      "geomean_change_pct": 22.11821238864453,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 512,
     "afd": {
      "median_us": 108.97799968719482,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 109.44015979766846,
      "paired_change_pct": [
       0.3643290030258095,
       0.4706231116011317,
       -0.0387786496916398
      ],
      "geomean_change_pct": 0.26515087321838937,
      "repeats_faster": 1
     },
     "masked": {
      "median_us": 118.52503776550293,
      "paired_change_pct": [
       8.760518733791645,
       8.189222114383377,
       9.112308001285975
      ],
      "geomean_change_pct": 8.686683543826845,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 2048,
     "afd": {
      "median_us": 366.67598724365234,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 368.4142303466797,
      "paired_change_pct": [
       0.301717048828265,
       0.49980280250361275,
       0.48961561342679705
      ],
      "geomean_change_pct": 0.4303371775690401,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 370.8937644958496,
      "paired_change_pct": [
       1.1657747684489461,
       1.0296455823651351,
       1.1659397426983187
      ],
      "geomean_change_pct": 1.1204329716654193,
      "repeats_faster": 0
     }
    }
   ],
   "pod": "afdep-v21-h100-c",
   "python": "3.12.3 (main, Aug 31 2026, 10:18:26) [GCC 13.3.0]"
  },
  "sa-ffnranks3attentionranks1experts96": {
   "arms": {
    "afd": {},
    "serial": {
     "_env": {
      "AFDEP_SERIAL_WAIT": "1"
     }
    },
    "masked": {
     "role_aware": false
    }
   },
   "passthrough": [
    "--payload",
    "bf16",
    "--routing",
    "random",
    "--hidden",
    "2048",
    "--ffn-ranks",
    "3",
    "--attention-ranks",
    "1",
    "--experts",
    "96",
    "--tokens",
    "1",
    "8",
    "32",
    "128",
    "512",
    "2048",
    "--capacity",
    "2048",
    "--warmup",
    "50",
    "--iterations",
    "200",
    "--blocks",
    "2"
   ],
   "processes": [
    {
     "order": [
      "afd",
      "serial",
      "masked"
     ],
     "median_us": {
      "afd": {
       "1": 19.8881596326828,
       "8": 20.822720527648926,
       "32": 23.01919937133789,
       "128": 28.538399934768677,
       "512": 59.07536029815674,
       "2048": 187.73856163024902
      },
      "serial": {
       "1": 20.134639739990234,
       "8": 21.14359974861145,
       "32": 23.093680143356323,
       "128": 29.499919414520264,
       "512": 60.61255931854248,
       "2048": 188.4061622619629
      },
      "masked": {
       "1": 30.6768798828125,
       "8": 31.156558990478516,
       "32": 32.925920486450195,
       "128": 38.90048027038574,
       "512": 68.12383890151978,
       "2048": 197.27176666259766
      }
     }
    },
    {
     "order": [
      "serial",
      "masked",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 19.71448004245758,
       "8": 21.108399629592896,
       "32": 22.931278944015503,
       "128": 29.322960376739502,
       "512": 60.48767805099487,
       "2048": 188.0455207824707
      },
      "masked": {
       "1": 31.110639572143555,
       "8": 31.062639951705933,
       "32": 32.94991970062256,
       "128": 39.372159242630005,
       "512": 68.14079999923706,
       "2048": 196.9767189025879
      },
      "afd": {
       "1": 20.124000310897827,
       "8": 21.014959812164307,
       "32": 23.169840574264526,
       "128": 28.550400733947754,
       "512": 59.30520057678223,
       "2048": 187.98824310302734
      }
     }
    },
    {
     "order": [
      "masked",
      "afd",
      "serial"
     ],
     "median_us": {
      "masked": {
       "1": 30.45184016227722,
       "8": 30.92168092727661,
       "32": 32.602399587631226,
       "128": 39.389519691467285,
       "512": 67.87295818328857,
       "2048": 197.81448364257812
      },
      "afd": {
       "1": 19.767199754714966,
       "8": 20.82271933555603,
       "32": 22.8547203540802,
       "128": 28.603839874267578,
       "512": 59.557199478149414,
       "2048": 187.7996063232422
      },
      "serial": {
       "1": 19.873040318489075,
       "8": 21.06559991836548,
       "32": 23.12079906463623,
       "128": 29.23143982887268,
       "512": 59.98920202255249,
       "2048": 188.03071975708008
      }
     }
    }
   ],
   "table": [
    {
     "tokens": 1,
     "afd": {
      "median_us": 19.8881596326828,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 19.873040318489075,
      "paired_change_pct": [
       1.239330897678359,
       -2.0349844072427126,
       0.5354352922389172
      ],
      "geomean_change_pct": -0.09670470630420702,
      "repeats_faster": 1
     },
     "masked": {
      "median_us": 30.6768798828125,
      "paired_change_pct": [
       54.24695119803984,
       54.59470826630872,
       54.05237231446354
      ],
      "geomean_change_pct": 54.29784756787128,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 8,
     "afd": {
      "median_us": 20.822720527648926,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 21.108399629592896,
      "paired_change_pct": [
       1.5410052713162647,
       0.44463476620355546,
       1.166421056229261
      ],
      "geomean_change_pct": 1.0496614854024644,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 31.062639951705933,
      "paired_change_pct": [
       49.627705703047134,
       47.81203594653378,
       48.49972488692198
      ],
      "geomean_change_pct": 48.644606255606185,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 32,
     "afd": {
      "median_us": 23.01919937133789,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 23.093680143356323,
      "paired_change_pct": [
       0.32355935068346664,
       -1.0296213713011149,
       1.1642177477289906
      ],
      "geomean_change_pct": 0.14863410502761187,
      "repeats_faster": 1
     },
     "masked": {
      "median_us": 32.925920486450195,
      "paired_change_pct": [
       43.03677532524242,
       42.210385932569075,
       42.65061695148151
      ],
      "geomean_change_pct": 42.63219311866322,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 128,
     "afd": {
      "median_us": 28.550400733947754,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 29.322960376739502,
      "paired_change_pct": [
       3.369212997047377,
       2.705950259651302,
       2.194110851423492
      ],
      "geomean_change_pct": 2.7552991881469646,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 39.372159242630005,
      "paired_change_pct": [
       36.30925475605527,
       37.90405118837676,
       37.70710458669102
      ],
      "geomean_change_pct": 37.30496390326603,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 512,
     "afd": {
      "median_us": 59.30520057678223,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 60.48767805099487,
      "paired_change_pct": [
       2.602098425853705,
       1.993884959012493,
       0.7253573844780536
      ],
      "geomean_change_pct": 1.7707710364111673,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 68.12383890151978,
      "paired_change_pct": [
       15.31684031666476,
       14.89852379980643,
       13.96264226324153
      ],
      "geomean_change_pct": 14.724603228373546,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 2048,
     "afd": {
      "median_us": 187.7996063232422,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 188.0455207824707,
      "paired_change_pct": [
       0.35560122860038845,
       0.030468756182777668,
       0.12306385426608113
      ],
      "geomean_change_pct": 0.16961794561369814,
      "repeats_faster": 0
     },
     "masked": {
      "median_us": 197.27176666259766,
      "paired_change_pct": [
       5.07791523998371,
       4.7814031618107045,
       5.332746705601843
      ],
      "geomean_change_pct": 5.063780104007698,
      "repeats_faster": 0
     }
    }
   ],
   "pod": "afdep-v21-h100-c",
   "python": "3.12.3 (main, Aug 31 2026, 10:18:26) [GCC 13.3.0]"
  },
  "sa5-ffnranks2attentionranks2": {
   "arms": {
    "afd": {},
    "serial": {
     "_env": {
      "AFDEP_SERIAL_WAIT": "1"
     }
    }
   },
   "passthrough": [
    "--payload",
    "bf16",
    "--routing",
    "random",
    "--hidden",
    "2048",
    "--ffn-ranks",
    "2",
    "--attention-ranks",
    "2",
    "--tokens",
    "1",
    "8",
    "32",
    "128",
    "--capacity",
    "2048",
    "--warmup",
    "50",
    "--iterations",
    "300",
    "--blocks",
    "3"
   ],
   "processes": [
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 20.384213129679363,
       "8": 24.66335932413737,
       "32": 25.93418598175049,
       "128": 35.146026611328125
      },
      "serial": {
       "1": 20.40671984354655,
       "8": 25.26250680287679,
       "32": 25.92565377553304,
       "128": 35.41674613952637
      }
     }
    },
    {
     "order": [
      "serial",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 20.689813296000164,
       "8": 24.618879954020183,
       "32": 26.251306533813477,
       "128": 35.50463994344076
      },
      "afd": {
       "1": 20.452906290690105,
       "8": 24.81162707010905,
       "32": 25.70133368174235,
       "128": 35.13792037963867
      }
     }
    },
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 20.53813298543294,
       "8": 25.02687931060791,
       "32": 25.95370610555013,
       "128": 35.38400014241537
      },
      "serial": {
       "1": 20.507307052612305,
       "8": 25.046292940775555,
       "32": 25.846932729085285,
       "128": 35.476907094319664
      }
     }
    },
    {
     "order": [
      "serial",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 20.947519938151043,
       "8": 24.67903931935628,
       "32": 26.186985969543457,
       "128": 36.20160102844238
      },
      "afd": {
       "1": 20.211520195007324,
       "8": 24.335360527038574,
       "32": 25.637226104736328,
       "128": 35.263681411743164
      }
     }
    },
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 20.185173352559406,
       "8": 24.011093775431316,
       "32": 25.736212730407715,
       "128": 35.09152094523112
      },
      "serial": {
       "1": 20.628479321797688,
       "8": 24.721066157023113,
       "32": 26.352747281392414,
       "128": 35.62272071838379
      }
     }
    }
   ],
   "table": [
    {
     "tokens": 1,
     "afd": {
      "median_us": 20.384213129679363,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 20.628479321797688,
      "paired_change_pct": [
       0.1104124732409506,
       1.1583048489196557,
       -0.1500912124899556,
       3.6414863208831028,
       2.196195997405548
      ],
      "geomean_change_pct": 1.3816525074406893,
      "repeats_faster": 1
     }
    },
    {
     "tokens": 8,
     "afd": {
      "median_us": 24.66335932413737,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 24.721066157023113,
      "paired_change_pct": [
       2.429301989502486,
       -0.7768419037745167,
       0.07757111834321151,
       1.4122609440523792,
       2.9568514797033485
      ],
      "geomean_change_pct": 1.2101216182632824,
      "repeats_faster": 1
     }
    },
    {
     "tokens": 32,
     "afd": {
      "median_us": 25.736212730407715,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 26.186985969543457,
      "paired_change_pct": [
       -0.03289945642964698,
       2.1398611405983647,
       -0.4113993432406593,
       2.144381231265746,
       2.395591602552516
      ],
      "geomean_change_pct": 1.2398623749362292,
      "repeats_faster": 2
     }
    },
    {
     "tokens": 128,
     "afd": {
      "median_us": 35.146026611328125,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 35.50463994344076,
      "paired_change_pct": [
       0.770270651621785,
       1.0436575637942092,
       0.2625676902847651,
       2.659732560953998,
       1.5137553427271966
      ],
      "geomean_change_pct": 1.2467440248723927,
      "repeats_faster": 0
     }
    }
   ],
   "pod": "afdep-v21-h100-c",
   "python": "3.12.3 (main, Aug 31 2026, 10:18:26) [GCC 13.3.0]"
  },
  "sa5-ffnranks1attentionranks3": {
   "arms": {
    "afd": {},
    "serial": {
     "_env": {
      "AFDEP_SERIAL_WAIT": "1"
     }
    }
   },
   "passthrough": [
    "--payload",
    "bf16",
    "--routing",
    "random",
    "--hidden",
    "2048",
    "--ffn-ranks",
    "1",
    "--attention-ranks",
    "3",
    "--tokens",
    "1",
    "8",
    "32",
    "128",
    "--capacity",
    "2048",
    "--warmup",
    "50",
    "--iterations",
    "300",
    "--blocks",
    "3"
   ],
   "processes": [
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 24.40106709798177,
       "8": 28.086398442586262,
       "32": 33.050880432128906,
       "128": 45.26325225830078
      },
      "serial": {
       "1": 24.645546277364094,
       "8": 27.122879028320312,
       "32": 33.14506530761719,
       "128": 45.74165344238281
      }
     }
    },
    {
     "order": [
      "serial",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 24.820906321207683,
       "8": 27.722558975219727,
       "32": 33.43776067097982,
       "128": 45.9552001953125
      },
      "afd": {
       "1": 24.47882652282715,
       "8": 27.329174677530926,
       "32": 32.930027643839516,
       "128": 45.14261245727539
      }
     }
    },
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 24.225813547770183,
       "8": 27.27349281311035,
       "32": 32.89482752482096,
       "128": 45.243307749430336
      },
      "serial": {
       "1": 25.01322587331136,
       "8": 27.87775993347168,
       "32": 33.978986740112305,
       "128": 46.499627431233726
      }
     }
    },
    {
     "order": [
      "serial",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 24.8853333791097,
       "8": 27.148160934448242,
       "32": 33.425280253092446,
       "128": 45.650132497151695
      },
      "afd": {
       "1": 24.50933297475179,
       "8": 27.335360844930012,
       "32": 32.77258555094401,
       "128": 45.04437446594238
      }
     }
    },
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 24.554346402486164,
       "8": 27.484585444132488,
       "32": 32.94272104899088,
       "128": 45.332587560017906
      },
      "serial": {
       "1": 25.348587036132812,
       "8": 27.700799306233723,
       "32": 33.03327878316244,
       "128": 45.709333419799805
      }
     }
    }
   ],
   "table": [
    {
     "tokens": 1,
     "afd": {
      "median_us": 24.47882652282715,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 24.8853333791097,
      "paired_change_pct": [
       1.001920032433934,
       1.3974517857771263,
       3.2503029216686574,
       1.5341111271581553,
       3.234623396720626
      ],
      "geomean_change_pct": 2.0791560505294626,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 8,
     "afd": {
      "median_us": 27.335360844930012,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 27.700799306233723,
      "paired_change_pct": [
       -3.430555242729183,
       1.4394298486160473,
       2.215583916962993,
       -0.6848269226944925,
       0.7866731791924986
      ],
      "geomean_change_pct": 0.045249659766177786,
      "repeats_faster": 2
     }
    },
    {
     "tokens": 32,
     "afd": {
      "median_us": 32.930027643839516,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 33.425280253092446,
      "paired_change_pct": [
       0.2849693389611607,
       1.5418542390300383,
       3.295834928677732,
       1.9915874538914302,
       0.27489451778097695
      ],
      "geomean_change_pct": 1.4714957720970423,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 128,
     "afd": {
      "median_us": 45.243307749430336,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 45.74165344238281,
      "paired_change_pct": [
       1.056930645089249,
       1.8000458852623469,
       2.776807762954081,
       1.344802849171134,
       0.8310707154827845
      ],
      "geomean_change_pct": 1.5596072139189232,
      "repeats_faster": 0
     }
    }
   ],
   "pod": "afdep-v21-h100-c",
   "python": "3.12.3 (main, Aug 31 2026, 10:18:26) [GCC 13.3.0]"
  },
  "sa5-ffnranks3attentionranks1experts96": {
   "arms": {
    "afd": {},
    "serial": {
     "_env": {
      "AFDEP_SERIAL_WAIT": "1"
     }
    }
   },
   "passthrough": [
    "--payload",
    "bf16",
    "--routing",
    "random",
    "--hidden",
    "2048",
    "--ffn-ranks",
    "3",
    "--attention-ranks",
    "1",
    "--experts",
    "96",
    "--tokens",
    "1",
    "8",
    "32",
    "128",
    "--capacity",
    "2048",
    "--warmup",
    "50",
    "--iterations",
    "300",
    "--blocks",
    "3"
   ],
   "processes": [
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 19.98528003692627,
       "8": 20.788052876790363,
       "32": 23.113172849019367,
       "128": 28.710826237996418
      },
      "serial": {
       "1": 19.587626457214355,
       "8": 21.048107147216797,
       "32": 23.007465998331707,
       "128": 29.144852956136067
      }
     }
    },
    {
     "order": [
      "serial",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 19.801173210144043,
       "8": 21.260159810384113,
       "32": 23.202560742696125,
       "128": 29.1323725382487
      },
      "afd": {
       "1": 20.047680536905926,
       "8": 20.990506807963055,
       "32": 23.145707448323567,
       "128": 28.886292775472004
      }
     }
    },
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 20.085652669270832,
       "8": 20.94645341237386,
       "32": 23.11968008677165,
       "128": 28.947092692057293
      },
      "serial": {
       "1": 20.108906428019207,
       "8": 21.44074598948161,
       "32": 23.213547070821125,
       "128": 29.421974817911785
      }
     }
    },
    {
     "order": [
      "serial",
      "afd"
     ],
     "median_us": {
      "serial": {
       "1": 19.92842674255371,
       "8": 21.279573440551758,
       "32": 23.086826006571453,
       "128": 29.395627975463867
      },
      "afd": {
       "1": 19.880320231119793,
       "8": 20.752746264139812,
       "32": 22.979200681050617,
       "128": 29.08757209777832
      }
     }
    },
    {
     "order": [
      "afd",
      "serial"
     ],
     "median_us": {
      "afd": {
       "1": 20.297493934631348,
       "8": 20.922346115112305,
       "32": 23.162026405334473,
       "128": 29.053548177083332
      },
      "serial": {
       "1": 19.909226099650066,
       "8": 21.3371737798055,
       "32": 23.204906781514484,
       "128": 29.096533457438152
      }
     }
    }
   ],
   "table": [
    {
     "tokens": 1,
     "afd": {
      "median_us": 20.047680536905926,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 19.909226099650066,
      "paired_change_pct": [
       -1.989732337886585,
       -1.2296052219511688,
       0.11577298050140516,
       0.24198056608069063,
       -1.9128855819920854
      ],
      "geomean_change_pct": -0.9595754415841928,
      "repeats_faster": 3
     }
    },
    {
     "tokens": 8,
     "afd": {
      "median_us": 20.922346115112305,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 21.279573440551758,
      "paired_change_pct": [
       1.2509794542459485,
       1.2846426476885275,
       2.3597912609671257,
       2.5385901687734247,
       1.9827014733952986
      ],
      "geomean_change_pct": 1.8819422748152403,
      "repeats_faster": 0
     }
    },
    {
     "tokens": 32,
     "afd": {
      "median_us": 23.11968008677165,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 23.202560742696125,
      "paired_change_pct": [
       -0.45734461200183096,
       0.24563213070714074,
       0.40600468387617106,
       0.4683597441645926,
       0.18513223078848728
      ],
      "geomean_change_pct": 0.16901215477640452,
      "repeats_faster": 1
     }
    },
    {
     "tokens": 128,
     "afd": {
      "median_us": 28.947092692057293,
      "paired_change_pct": [
       0.0,
       0.0,
       0.0,
       0.0,
       0.0
      ],
      "geomean_change_pct": 0.0,
      "repeats_faster": 0
     },
     "serial": {
      "median_us": 29.144852956136067,
      "paired_change_pct": [
       1.511717964999737,
       0.8518911190488421,
       1.6405175155458585,
       1.0590635638134893,
       0.14795191311167688
      ],
      "geomean_change_pct": 1.0408257233357032,
      "repeats_faster": 0
     }
    }
   ],
   "pod": "afdep-v21-h100-c",
   "python": "3.12.3 (main, Aug 31 2026, 10:18:26) [GCC 13.3.0]"
  }
 },
 "patch_sha256": "b1d615a7df85307a7808b9aede8c99424f0b0299e9d36812b759ebb02e341e95"
}