{ "meta": { "date": "2026-10-15", "gpu": "NVIDIA Blackwell Ultra / GB300 NVL72", "shape": { "T": 2048, "D": 7168, "E": 384, "K": 6, "defaultF": 3072, "layers": 1, "sharedExperts": 0 }, "theoreticalTFPerGPU": { "bf16": 2500, "mxfp8": 5000, "nvfp4": 15000 }, "theoreticalHBMGBps": 8000, "theoreticalNVLinkGBpsPerDirection": 900, "theoreticalNVLinkGBpsBidirectional": 1800, "peakNote": "Peak values are theoretical NVIDIA datasheet specifications, not benchmarked ceilings. Percentages compare measured useful throughput with the corresponding dense theoretical peak.", "timingNote": "CUDA graphs; maximum rank in each window, then median of five windows of 30 iterations. Forward, backward, and combined medians are reduced separately.", "bandwidthNote": "NVLink TX/RX are hardware counters from a separate collection of at least one second. HBM is an analytical useful-byte estimate, not a DRAM-counter measurement.", "placementNote": "Four separate allocations. EP64 training ran at a different site from EP32. Some allocations shared an NVLink fabric and overlapped in lifetime; timing-window overlap is unconfirmed. These observations do not establish uncontended peak performance.", "memoryNote": "One layer; training has one activation slot and capacity/scratch factors 8. VMM is off and no training layer triggered capacity recompute. Arena budgets and PyTorch allocation peaks are separate accounting views.", "source": "Author-supplied report_isolated_v6/report.html and cached measurements", "metricsExample": { "caseID": "training-ep64-mxfp8-mega-single_layer-F3072-i1", "forwardByteBuckets": { "inputs": 29360128, "routed_activations": 327155712, "routed_weights": 408748032, "routing_scores": 49152, "shared_work": 0 }, "forwardModelBytes": 765313024, "trainingModelBytes": 2295939072, "rank0TXMiBs": 239035.92260694358 } }, "cases": [ { "id": "inference-ep8-bf16-standard-single_layer-F3072-i1", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 1.421941332022349, "backwardMS": null, "layerMS": 1.4241407990455628, "windows": { "combined": [ 1.3855274637540183, 1.408019196987152, 1.4241407990455628, 1.434552534421285, 1.4325418591499328 ], "forward": [ 1.38317760626475, 1.405833613872528, 1.421941332022349, 1.432310402393341, 1.4303274631500245 ] }, "usefulTFPerGPU": 1139.9839390712232, "nvlinkTXGBps": 208.44604936541793, "nvlinkRXGBps": 208.44627592724788, "nvlinkMaxTXGBps": 213.552893896034, "nvlinkMaxRXGBps": 213.55297075836768, "analyticalHBMGBps": 4703.434270325753, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 3221225472, "arenaBytes": 3221225472, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 9.054837703704834, "reservedPeakGiB": 9.302734375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308424704, "forwardFlopsPerLogicalPayloadByteRank0": 5263.837873012922, "capacityRows": 98304 }, { "id": "inference-ep8-mxfp8-mega-single_layer-F3072-i1", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 0.798988801240921, "backwardMS": null, "layerMS": 0.8012810746828715, "windows": { "combined": [ 0.7824000040690104, 0.784909862279892, 0.8012810746828715, 0.8086943964163462, 0.8139530618985494 ], "forward": [ 0.7802730619907379, 0.7827658692995707, 0.798988801240921, 0.8064565300941468, 0.811788801352183 ] }, "usefulTFPerGPU": 2026.1275215198896, "nvlinkTXGBps": 276.6890197474898, "nvlinkRXGBps": 276.6860961793667, "nvlinkMaxTXGBps": 286.7868694071573, "nvlinkMaxRXGBps": 286.78656847517476, "analyticalHBMGBps": 4525.93897769931, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2485255680, "arenaBytes": 2485255680, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 11.404590606689453, "reservedPeakGiB": 11.646484375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308424704, "forwardFlopsPerLogicalPayloadByteRank0": 5263.837873012922, "capacityRows": 104320 }, { "id": "inference-ep8-mxfp8-staged-single_layer-F3072-i1", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 0.8730517347653707, "backwardMS": null, "layerMS": 0.8751968006292979, "windows": { "combined": [ 0.8711434622605642, 0.8751968006292979, 0.8718506654103597, 0.8857493261496227, 0.8816415965557098 ], "forward": [ 0.8689802626768748, 0.8730517347653707, 0.8696896076202393, 0.8835786680380503, 0.8794442653656006 ] }, "usefulTFPerGPU": 1855.0086525917905, "nvlinkTXGBps": 259.5692902309277, "nvlinkRXGBps": 259.57041223098975, "nvlinkMaxTXGBps": 267.45635738018, "nvlinkMaxRXGBps": 267.4556379421145, "analyticalHBMGBps": 4143.695732653937, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2123120640, "arenaBytes": 2123120640, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 11.067397117614746, "reservedPeakGiB": 11.294921875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308424704, "forwardFlopsPerLogicalPayloadByteRank0": 5263.837873012922, "capacityRows": 104320 }, { "id": "inference-ep8-nvfp4-mega-single_layer-F3072-i1", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 0.6131392021973928, "backwardMS": null, "layerMS": 0.6152672012646992, "windows": { "combined": [ 0.6170901318391164, 0.6152672012646992, 0.615340801080068, 0.6145258645216624, 0.6142975926399231 ], "forward": [ 0.6149653295675913, 0.6131392021973928, 0.6131978631019592, 0.6122912029425304, 0.6121589362621307 ] }, "usefulTFPerGPU": 2638.6871176471855, "nvlinkTXGBps": 323.9519371956487, "nvlinkRXGBps": 323.95180052141154, "nvlinkMaxTXGBps": 334.34675210827874, "nvlinkMaxRXGBps": 334.3480346967716, "analyticalHBMGBps": 3478.476934250311, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2084931584, "arenaBytes": 2084931584, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 9.647485256195068, "reservedPeakGiB": 9.908203125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308424704, "forwardFlopsPerLogicalPayloadByteRank0": 5263.837873012922, "capacityRows": 104320 }, { "id": "inference-ep8-bf16-standard-single_layer-F3072-i4", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 4.459767421086629, "backwardMS": null, "layerMS": 4.461881558100383, "windows": { "combined": [ 4.26582187016805, 4.444423468907674, 4.461881558100383, 4.539646959304809, 4.701834694544474 ], "forward": [ 4.263611745834351, 4.44223682085673, 4.459767421086629, 4.537533871332804, 4.699724849065145 ] }, "usefulTFPerGPU": 363.8594204592006, "nvlinkTXGBps": 68.39689713392404, "nvlinkRXGBps": 68.39094798541194, "nvlinkMaxTXGBps": 155.96387430114592, "nvlinkMaxRXGBps": 156.03344002908344, "analyticalHBMGBps": 1501.2394553233682, "receiveRows": [ 49152, 18169, 10151, 6716, 4875, 3752, 3007, 2482 ], "slotBytes": 0, "scratchBytes": 3221225472, "arenaBytes": 3221225472, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 9.054840564727783, "reservedPeakGiB": 9.302734375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 176103424, "forwardFlopsPerLogicalPayloadByteRank0": 9219.000976880496, "capacityRows": 98304 }, { "id": "inference-ep8-mxfp8-mega-single_layer-F3072-i4", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 2.5287647883097333, "backwardMS": null, "layerMS": 2.530909864107768, "windows": { "combined": [ 2.482048002878825, 2.5041855812072753, 2.530909864107768, 2.5344650586446127, 2.536190931002299 ], "forward": [ 2.479909340540568, 2.502034139633179, 2.5287647883097333, 2.5322954734166463, 2.5340319951375325 ] }, "usefulTFPerGPU": 641.4679799196794, "nvlinkTXGBps": 92.5619644494005, "nvlinkRXGBps": 92.39100093845998, "nvlinkMaxTXGBps": 264.03013287111077, "nvlinkMaxRXGBps": 162.07218341573738, "analyticalHBMGBps": 1432.9033599458044, "receiveRows": [ 49152, 18169, 10151, 6716, 4875, 3752, 3007, 2482 ], "slotBytes": 0, "scratchBytes": 2485255680, "arenaBytes": 2485255680, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 11.404593467712402, "reservedPeakGiB": 11.6484375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 176103424, "forwardFlopsPerLogicalPayloadByteRank0": 9219.000976880496, "capacityRows": 104320 }, { "id": "inference-ep8-mxfp8-staged-single_layer-F3072-i4", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 2.583837874730428, "backwardMS": null, "layerMS": 2.5860064109166463, "windows": { "combined": [ 2.4667840162913004, 2.5860064109166463, 2.592633597056071, 2.5711349566777546, 2.5887829383214314 ], "forward": [ 2.464631462097168, 2.583837874730428, 2.5904778639475503, 2.5689557154973346, 2.5866304000218707 ] }, "usefulTFPerGPU": 627.8010878219472, "nvlinkTXGBps": 88.9431997090149, "nvlinkRXGBps": 89.20287875227152, "nvlinkMaxTXGBps": 248.0211459588056, "nvlinkMaxRXGBps": 152.29438697632014, "analyticalHBMGBps": 1402.3744228516891, "receiveRows": [ 49152, 18169, 10151, 6716, 4875, 3752, 3007, 2482 ], "slotBytes": 0, "scratchBytes": 2123120640, "arenaBytes": 2123120640, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 11.067399978637695, "reservedPeakGiB": 11.296875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 176103424, "forwardFlopsPerLogicalPayloadByteRank0": 9219.000976880496, "capacityRows": 104320 }, { "id": "inference-ep8-nvfp4-mega-single_layer-F3072-i4", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 1.754744533697764, "backwardMS": null, "layerMS": 1.7569098631540934, "windows": { "combined": [ 1.7569098631540934, 1.7570240020751953, 1.7557568113009134, 1.7570111910502115, 1.7564544002215068 ], "forward": [ 1.754744533697764, 1.754861867427826, 1.7535903930664063, 1.7548234701156615, 1.7542805353800455 ] }, "usefulTFPerGPU": 924.0642744035912, "nvlinkTXGBps": 112.5728884534317, "nvlinkRXGBps": 113.16516004738232, "nvlinkMaxTXGBps": 355.94273535102866, "nvlinkMaxRXGBps": 149.89585135518774, "analyticalHBMGBps": 1218.1574097135626, "receiveRows": [ 49152, 18169, 10151, 6716, 4875, 3752, 3007, 2482 ], "slotBytes": 0, "scratchBytes": 2084931584, "arenaBytes": 2084931584, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 9.647488117218018, "reservedPeakGiB": 9.91015625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 176103424, "forwardFlopsPerLogicalPayloadByteRank0": 9219.000976880496, "capacityRows": 104320 }, { "id": "inference-ep8-bf16-standard-single_layer-F3072-i8", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 8.738711484273274, "backwardMS": null, "layerMS": 8.740808550516764, "windows": { "combined": [ 8.338570674260458, 8.483467753728231, 8.969963677724202, 8.740808550516764, 8.752563190460204 ], "forward": [ 8.336463991800944, 8.480732838312784, 8.967849604288737, 8.738711484273274, 8.75046609242757 ] }, "usefulTFPerGPU": 185.73769560391727, "nvlinkTXGBps": 35.93409234051396, "nvlinkRXGBps": 35.940662777168185, "nvlinkMaxTXGBps": 143.61558750663892, "nvlinkMaxRXGBps": 143.61560097759175, "analyticalHBMGBps": 131.485370187184, "receiveRows": [ 98304, 0, 0, 0, 0, 0, 0, 0 ], "slotBytes": 0, "scratchBytes": 3221225472, "arenaBytes": 3221225472, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 9.054843425750732, "reservedPeakGiB": 9.302734375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 0, "forwardFlopsPerLogicalPayloadByteRank0": null, "capacityRows": 98304 }, { "id": "inference-ep8-mxfp8-mega-single_layer-F3072-i8", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 4.727780310312907, "backwardMS": null, "layerMS": 4.729938125610351, "windows": { "combined": [ 4.657137044270834, 4.729938125610351, 4.749203157424927, 4.728469308217367, 4.879204241434733 ], "forward": [ 4.654977035522461, 4.727780310312907, 4.747053893407186, 4.726290146509807, 4.8770698547363285 ] }, "usefulTFPerGPU": 343.2386629113682, "nvlinkTXGBps": 49.94794480860118, "nvlinkRXGBps": 49.91085649635804, "nvlinkMaxTXGBps": 263.4150968923056, "nvlinkMaxRXGBps": 135.76692798876286, "analyticalHBMGBps": 161.80191023138258, "receiveRows": [ 98304, 0, 0, 0, 0, 0, 0, 0 ], "slotBytes": 0, "scratchBytes": 2485255680, "arenaBytes": 2485255680, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 11.404596328735352, "reservedPeakGiB": 11.6484375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 0, "forwardFlopsPerLogicalPayloadByteRank0": null, "capacityRows": 104320 }, { "id": "inference-ep8-mxfp8-staged-single_layer-F3072-i8", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 4.831414445241292, "backwardMS": null, "layerMS": 4.8335946559906, "windows": { "combined": [ 4.686046950022379, 4.813766400019328, 4.8335946559906, 4.8640938599904375, 5.056841595967611 ], "forward": [ 4.683907206853231, 4.8116213321685795, 4.831414445241292, 4.861926380793253, 5.054708242416382 ] }, "usefulTFPerGPU": 335.8779031824462, "nvlinkTXGBps": 48.172440880247386, "nvlinkRXGBps": 48.16926469338439, "nvlinkMaxTXGBps": 254.19059364381084, "nvlinkMaxRXGBps": 131.0850163963656, "analyticalHBMGBps": 158.33206515393175, "receiveRows": [ 98304, 0, 0, 0, 0, 0, 0, 0 ], "slotBytes": 0, "scratchBytes": 2123120640, "arenaBytes": 2123120640, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 11.067402839660645, "reservedPeakGiB": 11.296875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 0, "forwardFlopsPerLogicalPayloadByteRank0": null, "capacityRows": 104320 }, { "id": "inference-ep8-nvfp4-mega-single_layer-F3072-i8", "mode": "inference", "ep": 8, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 3.288329593340556, "backwardMS": null, "layerMS": 3.2904735724131267, "windows": { "combined": [ 3.291323733329773, 3.2906485319137575, 3.2904735724131267, 3.286622929573059, 3.289036822319031 ], "forward": [ 3.28919038772583, 3.288493855794271, 3.288329593340556, 3.2844810644785563, 3.286898120244344 ] }, "usefulTFPerGPU": 493.3933071212541, "nvlinkTXGBps": 61.79379023166012, "nvlinkRXGBps": 61.81008174736244, "nvlinkMaxTXGBps": 381.27118983581255, "nvlinkMaxRXGBps": 112.90603409133968, "analyticalHBMGBps": 176.12007853781364, "receiveRows": [ 98304, 0, 0, 0, 0, 0, 0, 0 ], "slotBytes": 0, "scratchBytes": 2084931584, "arenaBytes": 2084931584, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 9.647490978240967, "reservedPeakGiB": 9.91015625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 0, "forwardFlopsPerLogicalPayloadByteRank0": null, "capacityRows": 104320 }, { "id": "inference-ep16-bf16-standard-single_layer-F3072-i1", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 1.2925087928771972, "backwardMS": null, "layerMS": 1.294708263874054, "windows": { "combined": [ 1.2688458641370137, 1.2784629305203756, 1.294708263874054, 1.300603731473287, 1.3026581327120463 ], "forward": [ 1.2667146762212118, 1.2762762745221456, 1.2925087928771972, 1.2984021385510762, 1.300436266263326 ] }, "usefulTFPerGPU": 1253.9486177605258, "nvlinkTXGBps": 248.73436032125176, "nvlinkRXGBps": 248.7341576114828, "nvlinkMaxTXGBps": 254.773445846917, "nvlinkMaxRXGBps": 254.77498368285373, "analyticalHBMGBps": 2724.5201984306964, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 3221225472, "arenaBytes": 3221225472, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.101712703704834, "reservedPeakGiB": 6.353515625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 4917.334259661311, "capacityRows": 98304 }, { "id": "inference-ep16-mxfp8-mega-single_layer-F3072-i1", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 0.8593397378921509, "backwardMS": null, "layerMS": 0.8614869376023611, "windows": { "combined": [ 0.8366176029046376, 0.8337674677371979, 0.8826047976811727, 0.8967285275459289, 0.8614869376023611 ], "forward": [ 0.8344789326190949, 0.8315679967403412, 0.8805013358592987, 0.8945866644382476, 0.8593397378921509 ] }, "usefulTFPerGPU": 1884.5296046001831, "nvlinkTXGBps": 285.745484537377, "nvlinkRXGBps": 285.7456276606602, "nvlinkMaxTXGBps": 297.14879885858807, "nvlinkMaxRXGBps": 297.1538063047755, "analyticalHBMGBps": 2311.7670542315855, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2412113664, "arenaBytes": 2412113664, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.862129211425781, "reservedPeakGiB": 7.173828125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 4917.334259661311, "capacityRows": 101248 }, { "id": "inference-ep16-mxfp8-staged-single_layer-F3072-i1", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 0.8376106699307759, "backwardMS": null, "layerMS": 0.8397354702154796, "windows": { "combined": [ 0.8342346668243408, 0.8329642653465271, 0.8397354702154796, 0.8867871960004171, 0.8632554650306702 ], "forward": [ 0.8321066637833913, 0.8308298627535502, 0.8376106699307759, 0.8845813274383545, 0.8610688030719758 ] }, "usefulTFPerGPU": 1933.344124991414, "nvlinkTXGBps": 289.8240126798271, "nvlinkRXGBps": 289.8244177115723, "nvlinkMaxTXGBps": 299.2589595284742, "nvlinkMaxRXGBps": 299.25859475419514, "analyticalHBMGBps": 2371.6482042719454, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2060599296, "arenaBytes": 2060599296, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.534758567810059, "reservedPeakGiB": 6.802734375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 4917.334259661311, "capacityRows": 101248 }, { "id": "inference-ep16-nvfp4-mega-single_layer-F3072-i1", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 0.5232149302959442, "backwardMS": null, "layerMS": 0.5253386696179708, "windows": { "combined": [ 0.5259690701961517, 0.5242229282855988, 0.5238954722881317, 0.5256693343321482, 0.5253386696179708 ], "forward": [ 0.5238400022188823, 0.5221002658208211, 0.5217727979024251, 0.5235551953315735, 0.5232149302959442 ] }, "usefulTFPerGPU": 3090.3828935125152, "nvlinkTXGBps": 403.21162489135764, "nvlinkRXGBps": 403.21182099986953, "nvlinkMaxTXGBps": 417.5913124886281, "nvlinkMaxRXGBps": 417.59283129692875, "analyticalHBMGBps": 2376.3315974965785, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2023557632, "arenaBytes": 2023557632, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 5.808122634887695, "reservedPeakGiB": 6.099609375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 4917.334259661311, "capacityRows": 101248 }, { "id": "inference-ep16-bf16-standard-single_layer-F3072-i4", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 4.250103521347046, "backwardMS": null, "layerMS": 4.252232535680135, "windows": { "combined": [ 4.135118929545085, 4.252232535680135, 4.272587680816651, 4.1521919409434, 4.340621868769328 ], "forward": [ 4.132990924517314, 4.250103521347046, 4.270456520716349, 4.150060764948527, 4.338500245412191 ] }, "usefulTFPerGPU": 381.7988842955704, "nvlinkTXGBps": 78.67217363589842, "nvlinkRXGBps": 78.69846856772845, "nvlinkMaxTXGBps": 187.63675341792685, "nvlinkMaxRXGBps": 187.82802617355972, "analyticalHBMGBps": 829.5545425612033, "receiveRows": [ 49152, 27080, 19107, 14919, 12314, 10527, 9220, 8219, 7428, 6784, 6250, 5800, 5414, 5079, 4787, 4528 ], "slotBytes": 0, "scratchBytes": 3221225472, "arenaBytes": 3221225472, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.101715564727783, "reservedPeakGiB": 6.353515625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 264155136, "forwardFlopsPerLogicalPayloadByteRank0": 6146.000651253663, "capacityRows": 98304 }, { "id": "inference-ep16-mxfp8-mega-single_layer-F3072-i4", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 2.5254154602686563, "backwardMS": null, "layerMS": 2.5275594711303713, "windows": { "combined": [ 2.3024885336558025, 2.546122654279073, 2.5275594711303713, 2.547765342394511, 2.508531188964844 ], "forward": [ 2.300323216120402, 2.543951996167501, 2.5254154602686563, 2.545594644546509, 2.506390396753947 ] }, "usefulTFPerGPU": 642.3182743794914, "nvlinkTXGBps": 99.31339660470142, "nvlinkRXGBps": 99.21241282340834, "nvlinkMaxTXGBps": 289.96982635417913, "nvlinkMaxRXGBps": 188.2091543421219, "analyticalHBMGBps": 787.9367994096451, "receiveRows": [ 49152, 27080, 19107, 14919, 12314, 10527, 9220, 8219, 7428, 6784, 6250, 5800, 5414, 5079, 4787, 4528 ], "slotBytes": 0, "scratchBytes": 2412113664, "arenaBytes": 2412113664, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.8621320724487305, "reservedPeakGiB": 7.17578125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 264155136, "forwardFlopsPerLogicalPayloadByteRank0": 6146.000651253663, "capacityRows": 101248 }, { "id": "inference-ep16-mxfp8-staged-single_layer-F3072-i4", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 2.5881247838338215, "backwardMS": null, "layerMS": 2.5902944008509317, "windows": { "combined": [ 2.5229834715525308, 2.5804778575897216, 2.5902944008509317, 2.5918474594751992, 2.6020543972651162 ], "forward": [ 2.5208138545354206, 2.5782922665278116, 2.5881247838338215, 2.5896725257237754, 2.59989333152771 ] }, "usefulTFPerGPU": 626.7618218819715, "nvlinkTXGBps": 96.13089889746905, "nvlinkRXGBps": 96.06866744850414, "nvlinkMaxTXGBps": 283.3279608965275, "nvlinkMaxRXGBps": 183.8503680700181, "analyticalHBMGBps": 768.853578707408, "receiveRows": [ 49152, 27080, 19107, 14919, 12314, 10527, 9220, 8219, 7428, 6784, 6250, 5800, 5414, 5079, 4787, 4528 ], "slotBytes": 0, "scratchBytes": 2060599296, "arenaBytes": 2060599296, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.534761428833008, "reservedPeakGiB": 6.8046875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 264155136, "forwardFlopsPerLogicalPayloadByteRank0": 6146.000651253663, "capacityRows": 101248 }, { "id": "inference-ep16-nvfp4-mega-single_layer-F3072-i4", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 1.3669024070103963, "backwardMS": null, "layerMS": 1.3690399924914043, "windows": { "combined": [ 1.3545951962471008, 1.356050141652425, 1.3690399924914043, 1.3770047903060914, 1.3718592007954915 ], "forward": [ 1.3524661262830098, 1.3539125363032023, 1.3669024070103963, 1.3748544017473856, 1.3697162747383118 ] }, "usefulTFPerGPU": 1185.865750301077, "nvlinkTXGBps": 154.95967799630333, "nvlinkRXGBps": 154.59678659750176, "nvlinkMaxTXGBps": 506.71306576937394, "nvlinkMaxRXGBps": 237.2616740036165, "analyticalHBMGBps": 911.8644355510587, "receiveRows": [ 49152, 27080, 19107, 14919, 12314, 10527, 9220, 8219, 7428, 6784, 6250, 5800, 5414, 5079, 4787, 4528 ], "slotBytes": 0, "scratchBytes": 2023557632, "arenaBytes": 2023557632, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 5.8081254959106445, "reservedPeakGiB": 6.1015625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 264155136, "forwardFlopsPerLogicalPayloadByteRank0": 6146.000651253663, "capacityRows": 101248 }, { "id": "inference-ep16-bf16-standard-single_layer-F3072-i8", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 8.269385592142742, "backwardMS": null, "layerMS": 8.271519915262859, "windows": { "combined": [ 7.936585601170858, 8.115829308827719, 8.351893313725789, 8.337491162618, 8.271519915262859 ], "forward": [ 7.934434159596761, 8.113692792256673, 8.349775997797648, 8.335362243652344, 8.269385592142742 ] }, "usefulTFPerGPU": 196.27561252585187, "nvlinkTXGBps": 40.72615436279082, "nvlinkRXGBps": 40.718043448841044, "nvlinkMaxTXGBps": 173.13722910164103, "nvlinkMaxRXGBps": 173.1802777934128, "analyticalHBMGBps": 426.4583597859719, "receiveRows": [ 98304, 32835, 17288, 10967, 7705, 5775, 4525, 3663, 3041, 2574, 2213, 1929, 1699, 1511, 1355, 1224 ], "slotBytes": 0, "scratchBytes": 3221225472, "arenaBytes": 3221225472, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.101718425750732, "reservedPeakGiB": 6.353515625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 176304128, "forwardFlopsPerLogicalPayloadByteRank0": 9208.50609855261, "capacityRows": 98304 }, { "id": "inference-ep16-mxfp8-mega-single_layer-F3072-i8", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 4.7424736499786375, "backwardMS": null, "layerMS": 4.744611183802287, "windows": { "combined": [ 4.515010086695353, 4.743419742584228, 4.757050657272339, 4.744611183802287, 4.966227197647095 ], "forward": [ 4.512861839930216, 4.74126722017924, 4.754909865061442, 4.7424736499786375, 4.964093859990438 ] }, "usefulTFPerGPU": 342.177172163335, "nvlinkTXGBps": 52.987839826223485, "nvlinkRXGBps": 52.976142411613964, "nvlinkMaxTXGBps": 288.731501226083, "nvlinkMaxRXGBps": 162.61657075434888, "analyticalHBMGBps": 419.75138590892595, "receiveRows": [ 98304, 32835, 17288, 10967, 7705, 5775, 4525, 3663, 3041, 2574, 2213, 1929, 1699, 1511, 1355, 1224 ], "slotBytes": 0, "scratchBytes": 2412113664, "arenaBytes": 2412113664, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.86213493347168, "reservedPeakGiB": 7.17578125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 176304128, "forwardFlopsPerLogicalPayloadByteRank0": 9208.50609855261, "capacityRows": 101248 }, { "id": "inference-ep16-mxfp8-staged-single_layer-F3072-i8", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 4.987132740020752, "backwardMS": null, "layerMS": 4.989276806513469, "windows": { "combined": [ 4.810789330800374, 4.940997330347697, 5.02371621131897, 4.989276806513469, 5.118072589238484 ], "forward": [ 4.808634646733602, 4.9388469696044925, 5.02154769897461, 4.987132740020752, 5.115928490956624 ] }, "usefulTFPerGPU": 325.39738740663466, "nvlinkTXGBps": 50.96916279136098, "nvlinkRXGBps": 50.928045659290596, "nvlinkMaxTXGBps": 278.2436174756397, "nvlinkMaxRXGBps": 156.45182951267887, "analyticalHBMGBps": 399.167494054456, "receiveRows": [ 98304, 32835, 17288, 10967, 7705, 5775, 4525, 3663, 3041, 2574, 2213, 1929, 1699, 1511, 1355, 1224 ], "slotBytes": 0, "scratchBytes": 2060599296, "arenaBytes": 2060599296, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.534764289855957, "reservedPeakGiB": 6.8046875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 176304128, "forwardFlopsPerLogicalPayloadByteRank0": 9208.50609855261, "capacityRows": 101248 }, { "id": "inference-ep16-nvfp4-mega-single_layer-F3072-i8", "mode": "inference", "ep": 16, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 2.527577614784241, "backwardMS": null, "layerMS": 2.5297407944997152, "windows": { "combined": [ 2.477957344055176, 2.5462442715962728, 2.5083754539489744, 2.5297407944997152, 2.5758069435755413 ], "forward": [ 2.475827217102051, 2.5441365400950113, 2.50621120929718, 2.527577614784241, 2.573634131749471 ] }, "usefulTFPerGPU": 641.7644216426786, "nvlinkTXGBps": 83.29559132350111, "nvlinkRXGBps": 83.44649948493003, "nvlinkMaxTXGBps": 521.1723552981025, "nvlinkMaxRXGBps": 185.31321078391437, "analyticalHBMGBps": 493.48094584009783, "receiveRows": [ 98304, 32835, 17288, 10967, 7705, 5775, 4525, 3663, 3041, 2574, 2213, 1929, 1699, 1511, 1355, 1224 ], "slotBytes": 0, "scratchBytes": 2023557632, "arenaBytes": 2023557632, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 5.808128356933594, "reservedPeakGiB": 6.1015625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 176304128, "forwardFlopsPerLogicalPayloadByteRank0": 9208.50609855261, "capacityRows": 101248 }, { "id": "inference-ep16-bf16-standard-width-F1536-i1", "mode": "inference", "ep": 16, "category": "width", "F": 1536, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 0.8416778743267059, "backwardMS": null, "layerMS": 0.8437930623690287, "windows": { "combined": [ 0.8419754683971405, 0.8399210691452026, 0.8437930623690287, 0.8575935979684194, 0.8506431957085927 ], "forward": [ 0.8398634672164917, 0.8378111978371938, 0.8416778743267059, 0.8554826696713765, 0.8485578695933024 ] }, "usefulTFPerGPU": 962.0235756204709, "nvlinkTXGBps": 388.3047816561587, "nvlinkRXGBps": 388.3045412575723, "nvlinkMaxTXGBps": 401.18895325960233, "nvlinkMaxRXGBps": 401.18924359736553, "analyticalHBMGBps": 2212.052356486061, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2315255808, "arenaBytes": 2315255808, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 3.781400203704834, "reservedPeakGiB": 4.029296875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 2458.6671298306555, "capacityRows": 98304 }, { "id": "inference-ep16-mxfp8-mega-width-F1536-i1", "mode": "inference", "ep": 16, "category": "width", "F": 1536, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 0.7136469344298045, "backwardMS": null, "layerMS": 0.7157674690087636, "windows": { "combined": [ 0.7166581312815349, 0.715637336174647, 0.7136629283428192, 0.7167637288570404, 0.7157674690087636 ], "forward": [ 0.7145141363143921, 0.713509335120519, 0.7115392049153646, 0.7146453320980072, 0.7136469344298045 ] }, "usefulTFPerGPU": 1134.0957141683973, "nvlinkTXGBps": 349.05865855643134, "nvlinkRXGBps": 349.0590536869431, "nvlinkMaxTXGBps": 361.23517349278467, "nvlinkMaxRXGBps": 361.2357191234647, "analyticalHBMGBps": 1534.8051309475613, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 1614886656, "arenaBytes": 1614886656, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 3.8817386627197266, "reservedPeakGiB": 4.166015625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 2458.6671298306555, "capacityRows": 101248 }, { "id": "inference-ep16-mxfp8-staged-width-F1536-i1", "mode": "inference", "ep": 16, "category": "width", "F": 1536, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 0.6342133303483327, "backwardMS": null, "layerMS": 0.6363626698652903, "windows": { "combined": [ 0.6374474704265595, 0.6358431975046793, 0.6364191989103953, 0.6363626698652903, 0.6359658658504486 ], "forward": [ 0.6353269378344218, 0.6337109406789144, 0.6342933396498363, 0.6342133303483327, 0.6338453272978465 ] }, "usefulTFPerGPU": 1275.6072242827138, "nvlinkTXGBps": 390.7594625202316, "nvlinkRXGBps": 390.7590779974098, "nvlinkMaxTXGBps": 406.9829875499988, "nvlinkMaxRXGBps": 406.9835569019498, "analyticalHBMGBps": 1726.3168253294173, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 1438531584, "arenaBytes": 1438531584, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 3.7174978256225586, "reservedPeakGiB": 3.970703125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 2458.6671298306555, "capacityRows": 101248 }, { "id": "inference-ep16-nvfp4-mega-width-F1536-i1", "mode": "inference", "ep": 16, "category": "width", "F": 1536, "imbalance": 1.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 0.4908319999774297, "backwardMS": null, "layerMS": 0.49295893410841624, "windows": { "combined": [ 0.4946122666200002, 0.49321599702040353, 0.491825063029925, 0.4927978664636612, 0.49295893410841624 ], "forward": [ 0.4924896001815796, 0.49109866519769035, 0.4896831969420115, 0.49068479935328163, 0.4908319999774297 ] }, "usefulTFPerGPU": 1646.686494103528, "nvlinkTXGBps": 427.792337960492, "nvlinkRXGBps": 427.79279037312, "nvlinkMaxTXGBps": 444.89632530762407, "nvlinkMaxRXGBps": 444.89913471196206, "analyticalHBMGBps": 1474.7160740982063, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 1284741632, "arenaBytes": 1284741632, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 3.228200912475586, "reservedPeakGiB": 3.521484375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 2458.6671298306555, "capacityRows": 101248 }, { "id": "inference-ep16-bf16-standard-width-F3072-i1", "mode": "inference", "ep": 16, "category": "width", "F": 3072, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 1.2953098654747008, "backwardMS": null, "layerMS": 1.2974815924962362, "windows": { "combined": [ 1.2702176173528035, 1.2799925406773884, 1.3003946820894876, 1.2974815924962362, 1.3016597390174867 ], "forward": [ 1.2681002656618754, 1.2778090596199037, 1.2981898784637451, 1.2953098654747008, 1.299502925078074 ] }, "usefulTFPerGPU": 1251.2683395874146, "nvlinkTXGBps": 250.3003577226281, "nvlinkRXGBps": 250.30095995289483, "nvlinkMaxTXGBps": 257.491853331806, "nvlinkMaxRXGBps": 257.49204253178914, "analyticalHBMGBps": 2718.696616892646, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 3221225472, "arenaBytes": 3221225472, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.101721286773682, "reservedPeakGiB": 6.353515625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 4917.334259661311, "capacityRows": 98304 }, { "id": "inference-ep16-mxfp8-mega-width-F3072-i1", "mode": "inference", "ep": 16, "category": "width", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 0.846201604604721, "backwardMS": null, "layerMS": 0.848337060213089, "windows": { "combined": [ 0.8368064065774282, 0.8537792007128397, 0.848337060213089, 0.8884448011716207, 0.838404264052709 ], "forward": [ 0.8346730669339498, 0.85163946946462, 0.846201604604721, 0.8862666626771291, 0.8362634658813477 ] }, "usefulTFPerGPU": 1913.741263973783, "nvlinkTXGBps": 285.48877894802234, "nvlinkRXGBps": 285.48905139299836, "nvlinkMaxTXGBps": 294.5888374622292, "nvlinkMaxRXGBps": 294.58808774969935, "analyticalHBMGBps": 2347.6012229145713, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2412113664, "arenaBytes": 2412113664, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.862137794494629, "reservedPeakGiB": 7.17578125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 4917.334259661311, "capacityRows": 101248 }, { "id": "inference-ep16-mxfp8-staged-width-F3072-i1", "mode": "inference", "ep": 16, "category": "width", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 0.8438901364803314, "backwardMS": null, "layerMS": 0.8460533301035563, "windows": { "combined": [ 0.8289749344189962, 0.83310719927152, 0.8648437360922495, 0.8590496043364207, 0.8460533301035563 ], "forward": [ 0.8268256028493245, 0.8309514641761779, 0.8626751959323883, 0.8568885366121928, 0.8438901364803314 ] }, "usefulTFPerGPU": 1918.9069768087609, "nvlinkTXGBps": 289.34914405813015, "nvlinkRXGBps": 289.34843707135724, "nvlinkMaxTXGBps": 297.6472408093367, "nvlinkMaxRXGBps": 297.6476825597186, "analyticalHBMGBps": 2353.9380428373647, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2060599296, "arenaBytes": 2060599296, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 6.534767150878906, "reservedPeakGiB": 6.8046875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 4917.334259661311, "capacityRows": 101248 }, { "id": "inference-ep16-nvfp4-mega-width-F3072-i1", "mode": "inference", "ep": 16, "category": "width", "F": 3072, "imbalance": 1.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 0.5241813321908315, "backwardMS": null, "layerMS": 0.5263392011324565, "windows": { "combined": [ 0.5265354633331298, 0.5255850712458293, 0.5283082664012909, 0.5263392011324565, 0.5259071985880533 ], "forward": [ 0.5243626654148101, 0.5233973264694214, 0.5261557360490163, 0.5241813321908315, 0.5237653295199076 ] }, "usefulTFPerGPU": 3084.5083064208943, "nvlinkTXGBps": 406.9353620424653, "nvlinkRXGBps": 406.93562981740655, "nvlinkMaxTXGBps": 418.8934178068692, "nvlinkMaxRXGBps": 418.8998410243265, "analyticalHBMGBps": 2371.8143685935297, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 2023557632, "arenaBytes": 2023557632, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 5.808131217956543, "reservedPeakGiB": 6.1015625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 4917.334259661311, "capacityRows": 101248 }, { "id": "inference-ep16-bf16-standard-width-F6144-i1", "mode": "inference", "ep": 16, "category": "width", "F": 6144, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 2.457576521237691, "backwardMS": null, "layerMS": 2.459730132420858, "windows": { "combined": [ 2.286559987068176, 2.3979637225468955, 2.459730132420858, 2.4705024083455402, 2.497936002413432 ], "forward": [ 2.284380793571472, 2.395748249689738, 2.457576521237691, 2.4682069301605223, 2.495728023846944 ] }, "usefulTFPerGPU": 1320.0615925212571, "nvlinkTXGBps": 133.58634634979893, "nvlinkRXGBps": 133.58521152205776, "nvlinkMaxTXGBps": 135.01410266609534, "nvlinkMaxRXGBps": 135.01452465086612, "analyticalHBMGBps": 2784.59311195204, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 5033164800, "arenaBytes": 5033164800, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 10.742337703704834, "reservedPeakGiB": 11.001953125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 9834.668519322622, "capacityRows": 98304 }, { "id": "inference-ep16-mxfp8-mega-width-F6144-i1", "mode": "inference", "ep": 16, "category": "width", "F": 6144, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 1.3863103985786438, "backwardMS": null, "layerMS": 1.3885248064994813, "windows": { "combined": [ 1.3079008022944132, 1.3647146741549174, 1.3885248064994813, 1.3976288040479024, 1.3934816042582194 ], "forward": [ 1.3057557384173075, 1.3625333309173584, 1.3863103985786438, 1.3954175949096679, 1.3912501215934754 ] }, "usefulTFPerGPU": 2338.449598147106, "nvlinkTXGBps": 174.83285731936954, "nvlinkRXGBps": 174.8328294262533, "nvlinkMaxTXGBps": 177.74758630732993, "nvlinkMaxRXGBps": 177.74768755167312, "analyticalHBMGBps": 2720.5449800521164, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 4006567680, "arenaBytes": 4006567680, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 12.82291030883789, "reservedPeakGiB": 13.091796875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 9834.668519322622, "capacityRows": 101248 }, { "id": "inference-ep16-mxfp8-staged-width-F6144-i1", "mode": "inference", "ep": 16, "category": "width", "F": 6144, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 1.4726090590159098, "backwardMS": null, "layerMS": 1.4748095949490865, "windows": { "combined": [ 1.3858229398727417, 1.4411957343419393, 1.4748095949490865, 1.4818965355555216, 1.4754432042439778 ], "forward": [ 1.3836991985638936, 1.438970669110616, 1.4726090590159098, 1.4796330730120342, 1.473193593819936 ] }, "usefulTFPerGPU": 2201.6369346228, "nvlinkTXGBps": 167.11662679675163, "nvlinkRXGBps": 167.1163909260578, "nvlinkMaxTXGBps": 171.9290677972481, "nvlinkMaxRXGBps": 171.92724396310425, "analyticalHBMGBps": 2561.3775533718363, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 3304734720, "arenaBytes": 3304734720, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 12.169280052185059, "reservedPeakGiB": 12.427734375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 9834.668519322622, "capacityRows": 101248 }, { "id": "inference-ep16-nvfp4-mega-width-F6144-i1", "mode": "inference", "ep": 16, "category": "width", "F": 6144, "imbalance": 1.0, "precision": "nvfp4", "pipeline": "mega", "forwardMS": 0.7590922594070435, "backwardMS": null, "layerMS": 0.7612448036670685, "windows": { "combined": [ 0.7510986685752868, 0.7525269349416097, 0.7612448036670685, 0.7693151990572612, 0.7741205275058747 ], "forward": [ 0.7490005393822988, 0.7503925283749898, 0.7590922594070435, 0.7671520054340363, 0.7719829360644023 ] }, "usefulTFPerGPU": 4265.375947572416, "nvlinkTXGBps": 254.1540809395933, "nvlinkRXGBps": 254.15348842517892, "nvlinkMaxTXGBps": 263.356780410062, "nvlinkMaxRXGBps": 263.35200299412986, "analyticalHBMGBps": 3009.7909384246573, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 0, "scratchBytes": 3501189632, "arenaBytes": 3501189632, "savedBytes": [], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [], "allocatedPeakGiB": 10.967966079711914, "reservedPeakGiB": 11.275390625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330158080, "forwardFlopsPerLogicalPayloadByteRank0": 9834.668519322622, "capacityRows": 101248 }, { "id": "training-ep32-bf16-standard-single_layer-F3072-i1", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 1.2345802664756775, "backwardMS": 2.328651722272237, "layerMS": 3.5639146407445272, "windows": { "backward": [ 2.2864736318588257, 2.318894950548808, 2.328651722272237, 2.330574957529704, 2.3312640110651652 ], "combined": [ 3.5032629569371543, 3.6314112027486165, 3.557626660664876, 3.564465061823527, 3.5639146407445272 ], "forward": [ 1.2185269514719645, 1.3143221418062845, 1.2307264010111492, 1.235868791739146, 1.2345802664756775 ] }, "usefulTFPerGPU": 1366.6132342177868, "nvlinkTXGBps": 191.80039853763523, "nvlinkRXGBps": 191.8003848291913, "nvlinkMaxTXGBps": 193.07723349358076, "nvlinkMaxRXGBps": 193.07561429308592, "analyticalHBMGBps": 1634.7293073166588, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3821010944, "scratchBytes": 5838471168, "arenaBytes": 9659482112, "savedBytes": [ 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.652589321136475, "reservedPeakGiB": 16.705078125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 341454848, "forwardFlopsPerLogicalPayloadByteRank0": 4754.648081283063, "capacityRows": 98304 }, { "id": "training-ep32-mxfp8-mega-single_layer-F3072-i1", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 1.0575701355934144, "backwardMS": 1.5402634580930075, "layerMS": 2.5967349370320636, "windows": { "backward": [ 1.524954660733541, 1.5404661258061727, 1.5389568050702414, 1.5402634580930075, 1.5426463921864828 ], "combined": [ 2.567465583483378, 2.5967349370320636, 2.5941589434941608, 2.599227706591288, 2.6017706553141275 ], "forward": [ 1.0431306600570678, 1.0575701355934144, 1.0562890609105429, 1.060043728351593, 1.06003839969635 ] }, "usefulTFPerGPU": 1875.621898949273, "nvlinkTXGBps": 263.6965425845954, "nvlinkRXGBps": 263.69676332080996, "nvlinkMaxTXGBps": 266.30467881853895, "nvlinkMaxRXGBps": 266.30105632935386, "analyticalHBMGBps": 1356.3891784910772, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3707174912, "scratchBytes": 5408378880, "arenaBytes": 9115553792, "savedBytes": [ 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.912933349609375, "reservedPeakGiB": 17.193359375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 341454848, "forwardFlopsPerLogicalPayloadByteRank0": 4754.648081283063, "capacityRows": 99712 }, { "id": "training-ep32-mxfp8-staged-single_layer-F3072-i1", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 0.9140789330005645, "backwardMS": 1.6219061334927878, "layerMS": 2.5343552112579344, "windows": { "backward": [ 1.5956213355064393, 1.639949866135915, 1.6218250513076782, 1.6219061334927878, 1.6300501346588134 ], "combined": [ 2.5054805199305217, 2.5602709452311196, 2.5343552112579344, 2.5327914714813233, 2.544157886505127 ], "forward": [ 0.9109802663326263, 0.9219413280487061, 0.9140789330005645, 0.9121983965237935, 0.9156330704689026 ] }, "usefulTFPerGPU": 1921.7877951869727, "nvlinkTXGBps": 270.0029009383232, "nvlinkRXGBps": 270.00349811408336, "nvlinkMaxTXGBps": 273.8807523828198, "nvlinkMaxRXGBps": 273.88269509969723, "analyticalHBMGBps": 1389.7748635842386, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3707174912, "scratchBytes": 5063774208, "arenaBytes": 8770949120, "savedBytes": [ 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.591995239257812, "reservedPeakGiB": 16.861328125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 341454848, "forwardFlopsPerLogicalPayloadByteRank0": 4754.648081283063, "capacityRows": 99712 }, { "id": "training-ep32-bf16-standard-single_layer-F3072-i4", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 4.242783959706625, "backwardMS": 8.255635166168213, "layerMS": 12.487484741210938, "windows": { "backward": [ 7.998866081237793, 8.305694802602131, 8.255635166168213, 8.277395184834798, 8.244460773468017 ], "combined": [ 12.103730138142904, 12.566025543212891, 12.487484741210938, 12.52098035812378, 12.480435276031494 ], "forward": [ 4.111116846402486, 4.266679477691651, 4.238100242614746, 4.250705083211263, 4.242783959706625 ] }, "usefulTFPerGPU": 390.02993914302857, "nvlinkTXGBps": 55.749762068631306, "nvlinkRXGBps": 55.746589338537284, "nvlinkMaxTXGBps": 135.64067661994892, "nvlinkMaxRXGBps": 135.6635150300525, "analyticalHBMGBps": 466.5499764554697, "receiveRows": [ 49152, 32274, 25234, 21191, 18507, 16569, 15089, 13914, 12954, 12152, 11469, 10879, 10363, 9908, 9501, 9136, 8806, 8506, 8231, 7979, 7746, 7531, 7330, 7143, 6969, 6805, 6651, 6505, 6368, 6239, 6116, 5999 ], "slotBytes": 3821010944, "scratchBytes": 5838471168, "arenaBytes": 9659482112, "savedBytes": [ 1484783616, 1035423744, 847990784, 740349952, 668891136, 617293824, 577890304, 546607104, 521048064, 499695616, 481511424, 465803264, 452065280, 439951360, 429115392, 419397632, 410611712, 402624512, 395302912, 388593664, 382390272, 376666112, 371314688, 366336000, 361703424, 357337088, 353236992, 349349888, 345702400, 342267904, 338993152, 335878144 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.652592182159424, "reservedPeakGiB": 16.705078125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308166656, "forwardFlopsPerLogicalPayloadByteRank0": 5268.245627093413, "capacityRows": 98304 }, { "id": "training-ep32-mxfp8-mega-single_layer-F3072-i4", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 3.519040012359619, "backwardMS": 4.592920509974162, "layerMS": 8.107110500335693, "windows": { "backward": [ 4.557900842030843, 4.560252777735392, 4.592920509974162, 4.608843739827474, 4.622974904378255 ], "combined": [ 8.013799476623536, 8.04951999982198, 8.107110500335693, 8.151750500996908, 8.168087418874105 ], "forward": [ 3.4611498514811196, 3.494921604792277, 3.519040012359619, 3.5482293367385864, 3.550376542409261 ] }, "usefulTFPerGPU": 600.7680434924781, "nvlinkTXGBps": 85.7151099258847, "nvlinkRXGBps": 85.71492155153844, "nvlinkMaxTXGBps": 209.7382594329818, "nvlinkMaxRXGBps": 209.71180132329565, "analyticalHBMGBps": 434.4560454497513, "receiveRows": [ 49152, 32274, 25234, 21191, 18507, 16569, 15089, 13914, 12954, 12152, 11469, 10879, 10363, 9908, 9501, 9136, 8806, 8506, 8231, 7979, 7746, 7531, 7330, 7143, 6969, 6805, 6651, 6505, 6368, 6239, 6116, 5999 ], "slotBytes": 3707174912, "scratchBytes": 5408378880, "arenaBytes": 9115553792, "savedBytes": [ 1299185664, 948240384, 772767744, 667484160, 632389632, 562200576, 527106048, 527106048, 492011520, 456916992, 456916992, 456916992, 421822464, 421822464, 421822464, 386727936, 386727936, 386727936, 386727936, 386727936, 386727936, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 316538880, 316538880 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.912936210632324, "reservedPeakGiB": 17.1953125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308166656, "forwardFlopsPerLogicalPayloadByteRank0": 5268.245627093413, "capacityRows": 99712 }, { "id": "training-ep32-mxfp8-staged-single_layer-F3072-i4", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 2.9127861420313517, "backwardMS": 5.193706639607748, "layerMS": 8.102012761433919, "windows": { "backward": [ 5.151729075113932, 5.161253341039022, 5.193706639607748, 5.210190916061402, 5.227560504277547 ], "combined": [ 8.041403802235921, 8.056411711374919, 8.102012761433919, 8.128940804799397, 8.162324301401775 ], "forward": [ 2.8945653279622396, 2.9003423929214476, 2.9127861420313517, 2.9230613390604656, 2.939476251602173 ] }, "usefulTFPerGPU": 601.1460432212409, "nvlinkTXGBps": 85.92608581611664, "nvlinkRXGBps": 85.93096447409705, "nvlinkMaxTXGBps": 210.48163636458634, "nvlinkMaxRXGBps": 210.5425847766518, "analyticalHBMGBps": 434.7294026449587, "receiveRows": [ 49152, 32274, 25234, 21191, 18507, 16569, 15089, 13914, 12954, 12152, 11469, 10879, 10363, 9908, 9501, 9136, 8806, 8506, 8231, 7979, 7746, 7531, 7330, 7143, 6969, 6805, 6651, 6505, 6368, 6239, 6116, 5999 ], "slotBytes": 3707174912, "scratchBytes": 5063774208, "arenaBytes": 8770949120, "savedBytes": [ 1299185664, 948240384, 772767744, 667484160, 632389632, 562200576, 527106048, 527106048, 492011520, 456916992, 456916992, 456916992, 421822464, 421822464, 421822464, 386727936, 386727936, 386727936, 386727936, 386727936, 386727936, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 316538880, 316538880 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.591998100280762, "reservedPeakGiB": 16.86328125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308166656, "forwardFlopsPerLogicalPayloadByteRank0": 5268.245627093413, "capacityRows": 99712 }, { "id": "training-ep32-bf16-standard-single_layer-F3072-i8", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 8.295095459620159, "backwardMS": 16.21299107869466, "layerMS": 24.511511611938477, "windows": { "backward": [ 15.841358884175618, 16.21299107869466, 16.21982494990031, 16.239160569508872, 16.201988156636556 ], "combined": [ 23.89958292643229, 24.516086387634278, 24.510075823465982, 24.511511611938477, 24.55817279815674 ], "forward": [ 8.062793588638305, 8.30770557721456, 8.295095459620159, 8.27706880569458, 8.360834217071533 ] }, "usefulTFPerGPU": 198.70226654205192, "nvlinkTXGBps": 28.409838678296936, "nvlinkRXGBps": 28.431198669081493, "nvlinkMaxTXGBps": 124.13271765415574, "nvlinkMaxRXGBps": 124.7623129908934, "analyticalHBMGBps": 237.68569659173508, "receiveRows": [ 98304, 48816, 32414, 24242, 19351, 16096, 13776, 12038, 10688, 9609, 8727, 7993, 7373, 6841, 6381, 5978, 5623, 5308, 5025, 4772, 4542, 4334, 4144, 3969, 3809, 3661, 3524, 3397, 3279, 3168, 3065, 2969 ], "slotBytes": 3821010944, "scratchBytes": 5838471168, "arenaBytes": 9659482112, "savedBytes": [ 2793406464, 1475837952, 1039151104, 821579776, 691361792, 604700672, 542932992, 496660480, 460718080, 431990784, 408508416, 388966400, 372459520, 358295552, 346048512, 335319040, 325867520, 317480960, 309946368, 303210496, 297086976, 291549184, 286490624, 281831424, 277571584, 273631232, 269983744, 266602496, 263460864, 260505600, 257763328, 255207424 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.652595043182373, "reservedPeakGiB": 16.705078125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 264327168, "forwardFlopsPerLogicalPayloadByteRank0": 6142.000650829808, "capacityRows": 98304 }, { "id": "training-ep32-mxfp8-mega-single_layer-F3072-i8", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 6.848523759841919, "backwardMS": 8.870887406667073, "layerMS": 15.713252194722493, "windows": { "backward": [ 8.743787733713786, 8.97035083770752, 8.870887406667073, 8.85702298482259, 8.877724774678548 ], "combined": [ 15.415010134379068, 15.855515607198079, 15.713252194722493, 15.687073040008546, 15.724353059132893 ], "forward": [ 6.676634645462036, 6.890892775853475, 6.848523759841919, 6.836436287562052, 6.852634684244792 ] }, "usefulTFPerGPU": 309.9608440892854, "nvlinkTXGBps": 44.25455830589854, "nvlinkRXGBps": 44.270591044340996, "nvlinkMaxTXGBps": 193.99887413706816, "nvlinkMaxRXGBps": 194.21638735413873, "analyticalHBMGBps": 224.15367133119474, "receiveRows": [ 98304, 48816, 32414, 24242, 19351, 16096, 13776, 12038, 10688, 9609, 8727, 7993, 7373, 6841, 6381, 5978, 5623, 5308, 5025, 4772, 4542, 4334, 4144, 3969, 3809, 3661, 3524, 3397, 3279, 3168, 3065, 2969 ], "slotBytes": 3707174912, "scratchBytes": 5408378880, "arenaBytes": 9115553792, "savedBytes": [ 2422210560, 1299185664, 948240384, 737673216, 632389632, 562200576, 492011520, 456916992, 421822464, 421822464, 386727936, 386727936, 351633408, 351633408, 351633408, 316538880, 316538880, 316538880, 316538880, 316538880, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 246349824, 246349824 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.912939071655273, "reservedPeakGiB": 17.1953125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 264327168, "forwardFlopsPerLogicalPayloadByteRank0": 6142.000650829808, "capacityRows": 99712 }, { "id": "training-ep32-mxfp8-staged-single_layer-F3072-i8", "mode": "training", "ep": 32, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 5.752847973505656, "backwardMS": 10.25458459854126, "layerMS": 16.00031461715698, "windows": { "backward": [ 10.147498734792073, 10.291661898295084, 10.25458459854126, 10.255795160929361, 10.198358472188314 ], "combined": [ 15.847154172261556, 16.061930720011393, 16.00031461715698, 16.002808570861816, 15.923786735534668 ], "forward": [ 5.706554635365804, 5.777070887883505, 5.752847973505656, 5.755426152547201, 5.732945092519125 ] }, "usefulTFPerGPU": 304.3998215160981, "nvlinkTXGBps": 43.60685709239489, "nvlinkRXGBps": 43.59874361438779, "nvlinkMaxTXGBps": 191.04083465231258, "nvlinkMaxRXGBps": 190.83442577099422, "analyticalHBMGBps": 220.13211941615182, "receiveRows": [ 98304, 48816, 32414, 24242, 19351, 16096, 13776, 12038, 10688, 9609, 8727, 7993, 7373, 6841, 6381, 5978, 5623, 5308, 5025, 4772, 4542, 4334, 4144, 3969, 3809, 3661, 3524, 3397, 3279, 3168, 3065, 2969 ], "slotBytes": 3707174912, "scratchBytes": 5063774208, "arenaBytes": 8770949120, "savedBytes": [ 2422210560, 1299185664, 948240384, 737673216, 632389632, 562200576, 492011520, 456916992, 421822464, 421822464, 386727936, 386727936, 351633408, 351633408, 351633408, 316538880, 316538880, 316538880, 316538880, 316538880, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 281444352, 246349824, 246349824 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 13.592000961303711, "reservedPeakGiB": 16.86328125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 264327168, "forwardFlopsPerLogicalPayloadByteRank0": 6142.000650829808, "capacityRows": 99712 }, { "id": "training-ep64-bf16-standard-single_layer-F3072-i1", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 1.3662389357884726, "backwardMS": 2.467614928881327, "layerMS": 3.831937074661255, "windows": { "backward": [ 2.404980270067851, 2.479473042488098, 2.481828284263611, 2.467614928881327, 2.4462549368540447 ], "combined": [ 3.7418016036351522, 3.9473130623499553, 3.8543669462203978, 3.831937074661255, 3.8057173331578573 ], "forward": [ 1.3388821283976238, 1.4697152018547057, 1.3745386759440104, 1.3662389357884726, 1.3618677337964376 ] }, "usefulTFPerGPU": 1271.0263291822332, "nvlinkTXGBps": 175.265420136353, "nvlinkRXGBps": 175.266038743109, "nvlinkMaxTXGBps": 176.35602481379496, "nvlinkMaxRXGBps": 176.35656239244017, "analyticalHBMGBps": 899.7708670111168, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3821010944, "scratchBytes": 5838471168, "arenaBytes": 9659482112, "savedBytes": [ 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 11.437745571136475, "reservedPeakGiB": 13.013671875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 4681.529888383629, "capacityRows": 98304 }, { "id": "training-ep64-mxfp8-mega-single_layer-F3072-i1", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 1.1498677333196003, "backwardMS": 1.4917216062545777, "layerMS": 2.6411285400390625, "windows": { "backward": [ 1.4739189227422078, 1.49749546845754, 1.4917216062545777, 1.4960277318954467, 1.4873279929161072 ], "combined": [ 2.5679616133371987, 2.656395745277405, 2.6440746545791627, 2.6411285400390625, 2.6361962795257567 ], "forward": [ 1.0946549336115519, 1.1600202639897665, 1.1533130725224814, 1.145966931184133, 1.1498677333196003 ] }, "usefulTFPerGPU": 1844.0953705312522, "nvlinkTXGBps": 259.80531104501244, "nvlinkRXGBps": 259.8061210390631, "nvlinkMaxTXGBps": 264.2892762487284, "nvlinkMaxRXGBps": 264.28863633785403, "analyticalHBMGBps": 869.3022839267198, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3566796800, "scratchBytes": 5366722560, "arenaBytes": 8933519360, "savedBytes": [ 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 11.153888702392578, "reservedPeakGiB": 12.798828125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 4681.529888383629, "capacityRows": 98944 }, { "id": "training-ep64-mxfp8-staged-single_layer-F3072-i1", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 0.9781514664491018, "backwardMS": 1.6477727850278219, "layerMS": 2.624785033861796, "windows": { "backward": [ 1.607476270198822, 1.642354150613149, 1.6627935965855916, 1.6687402685483297, 1.6477727850278219 ], "combined": [ 2.5638879855473835, 2.614733862876892, 2.6441983938217164, 2.64934827486674, 2.624785033861796 ], "forward": [ 0.9573503990968069, 0.9732768019040425, 0.9832640051841736, 0.9821738680203755, 0.9781514664491018 ] }, "usefulTFPerGPU": 1855.5778285957906, "nvlinkTXGBps": 265.1536810190743, "nvlinkRXGBps": 265.1541203532544, "nvlinkMaxTXGBps": 278.39907512250585, "nvlinkMaxRXGBps": 278.3998833961057, "analyticalHBMGBps": 874.7150880474309, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3566796800, "scratchBytes": 5024772096, "arenaBytes": 8591568896, "savedBytes": [ 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 10.83542251586914, "reservedPeakGiB": 12.486328125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 4681.529888383629, "capacityRows": 98944 }, { "id": "training-ep64-bf16-standard-single_layer-F3072-i4", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 4.722298653920491, "backwardMS": 9.123507181803385, "layerMS": 13.84099629720052, "windows": { "backward": [ 8.74072535832723, 9.206537659962972, 9.254322147369384, 9.123507181803385, 9.032516225179036 ], "combined": [ 13.259110355377198, 13.97249813079834, 14.047989336649577, 13.84099629720052, 13.694055493672689 ], "forward": [ 4.5229290644327795, 4.771149826049805, 4.798165353139241, 4.722298653920491, 4.665635220209757 ] }, "usefulTFPerGPU": 351.88889651311484, "nvlinkTXGBps": 50.73755589171379, "nvlinkRXGBps": 50.73522644446934, "nvlinkMaxTXGBps": 125.53312514670492, "nvlinkMaxRXGBps": 125.5332737207417, "analyticalHBMGBps": 249.10528620669922, "receiveRows": [ 49152, 35589, 29463, 25768, 23224, 21333, 19855, 18657, 17661, 16815, 16085, 15446, 14881, 14376, 13921, 13509, 13133, 12788, 12469, 12175, 11901, 11646, 11408, 11184, 10973, 10774, 10587, 10409, 10240, 10080, 9927, 9781, 9642, 9509, 9381, 9259, 9141, 9029, 8920, 8815, 8715, 8617, 8523, 8433, 8345, 8260, 8177, 8098, 8020, 7945, 7872, 7801, 7732, 7665, 7600, 7536, 7475, 7414, 7355, 7298, 7242, 7187, 7134, 7082 ], "slotBytes": 3821010944, "scratchBytes": 5838471168, "arenaBytes": 9659482112, "savedBytes": [ 1484783616, 1123682304, 960583680, 862208000, 794476544, 744130560, 704780288, 672884736, 646367232, 623843328, 604407808, 587395072, 572352512, 558907392, 546793472, 535824384, 525813760, 516628480, 508135424, 500307968, 493012992, 486223872, 479887360, 473923584, 468305920, 463007744, 458029056, 453289984, 448790528, 444530688, 440457216, 436570112, 432869376, 429328384, 425920512, 422672384, 419530752, 416548864, 413646848, 410851328, 408188928, 405579776, 403077120, 400680960, 398338048, 396075008, 393865216, 391761920, 389685248, 387688448, 385744896, 383854592, 382017536, 380233728, 378503168, 376799232, 375175168, 373551104, 371980288, 370462720, 368971776, 367507456, 366096384, 364711936 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 11.437748432159424, "reservedPeakGiB": 13.013671875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330301440, "forwardFlopsPerLogicalPayloadByteRank0": 4915.2, "capacityRows": 98304 }, { "id": "training-ep64-mxfp8-mega-single_layer-F3072-i4", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 3.7936405181884765, "backwardMS": 5.035453875859578, "layerMS": 8.816842778523762, "windows": { "backward": [ 4.944114112854004, 5.02805864016215, 5.141774892807007, 5.104949331283569, 5.035453875859578 ], "combined": [ 8.53193925221761, 8.816842778523762, 8.963053862253824, 8.93258876800537, 8.797443262736003 ], "forward": [ 3.592675232887268, 3.7936405181884765, 3.8258613189061483, 3.8319530804951984, 3.7660533507664997 ] }, "usefulTFPerGPU": 552.4078217122848, "nvlinkTXGBps": 78.80920472163075, "nvlinkRXGBps": 78.80903670674901, "nvlinkMaxTXGBps": 194.6885765272081, "nvlinkMaxRXGBps": 194.59682421993494, "analyticalHBMGBps": 260.4037669348594, "receiveRows": [ 49152, 35589, 29463, 25768, 23224, 21333, 19855, 18657, 17661, 16815, 16085, 15446, 14881, 14376, 13921, 13509, 13133, 12788, 12469, 12175, 11901, 11646, 11408, 11184, 10973, 10774, 10587, 10409, 10240, 10080, 9927, 9781, 9642, 9509, 9381, 9259, 9141, 9029, 8920, 8815, 8715, 8617, 8523, 8433, 8345, 8260, 8177, 8098, 8020, 7945, 7872, 7801, 7732, 7665, 7600, 7536, 7475, 7414, 7355, 7298, 7242, 7187, 7134, 7082 ], "slotBytes": 3566796800, "scratchBytes": 5366722560, "arenaBytes": 8933519360, "savedBytes": [ 1299185664, 1000882176, 860504064, 772767744, 720125952, 667484160, 632389632, 614842368, 579747840, 562200576, 544653312, 544653312, 527106048, 509558784, 509558784, 492011520, 492011520, 474464256, 474464256, 456916992, 456916992, 456916992, 439369728, 439369728, 439369728, 439369728, 421822464, 421822464, 421822464, 421822464, 404275200, 404275200, 404275200, 404275200, 404275200, 404275200, 386727936, 386727936, 386727936, 386727936, 386727936, 386727936, 386727936, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 11.153891563415527, "reservedPeakGiB": 12.80078125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330301440, "forwardFlopsPerLogicalPayloadByteRank0": 4915.2, "capacityRows": 98944 }, { "id": "training-ep64-mxfp8-staged-single_layer-F3072-i4", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 4.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 3.187670381863912, "backwardMS": 5.702828820546468, "layerMS": 8.88519681294759, "windows": { "backward": [ 5.598797845840454, 5.702828820546468, 5.7800490697224935, 5.661046409606934, 5.738124799728394 ], "combined": [ 8.732396825154622, 8.88519681294759, 8.99738556543986, 8.823900826772054, 8.938323179880777 ], "forward": [ 3.1389450788497926, 3.187670381863912, 3.2231082677841187, 3.1677813371022543, 3.205237356821696 ] }, "usefulTFPerGPU": 548.1581349516842, "nvlinkTXGBps": 78.85054280257322, "nvlinkRXGBps": 78.85045864280059, "nvlinkMaxTXGBps": 194.79284608840018, "nvlinkMaxRXGBps": 194.70221506287825, "analyticalHBMGBps": 258.4004744446782, "receiveRows": [ 49152, 35589, 29463, 25768, 23224, 21333, 19855, 18657, 17661, 16815, 16085, 15446, 14881, 14376, 13921, 13509, 13133, 12788, 12469, 12175, 11901, 11646, 11408, 11184, 10973, 10774, 10587, 10409, 10240, 10080, 9927, 9781, 9642, 9509, 9381, 9259, 9141, 9029, 8920, 8815, 8715, 8617, 8523, 8433, 8345, 8260, 8177, 8098, 8020, 7945, 7872, 7801, 7732, 7665, 7600, 7536, 7475, 7414, 7355, 7298, 7242, 7187, 7134, 7082 ], "slotBytes": 3566796800, "scratchBytes": 5024772096, "arenaBytes": 8591568896, "savedBytes": [ 1299185664, 1000882176, 860504064, 772767744, 720125952, 667484160, 632389632, 614842368, 579747840, 562200576, 544653312, 544653312, 527106048, 509558784, 509558784, 492011520, 492011520, 474464256, 474464256, 456916992, 456916992, 456916992, 439369728, 439369728, 439369728, 439369728, 421822464, 421822464, 421822464, 421822464, 404275200, 404275200, 404275200, 404275200, 404275200, 404275200, 386727936, 386727936, 386727936, 386727936, 386727936, 386727936, 386727936, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 369180672, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408, 351633408 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 10.83542537689209, "reservedPeakGiB": 12.48828125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 330301440, "forwardFlopsPerLogicalPayloadByteRank0": 4915.2, "capacityRows": 98944 }, { "id": "training-ep64-bf16-standard-single_layer-F3072-i8", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 9.244633642832438, "backwardMS": 18.143260701497397, "layerMS": 27.38311570485433, "windows": { "backward": [ 17.64665288925171, 18.142543919881184, 18.17554327646891, 18.2386048634847, 18.143260701497397 ], "combined": [ 26.597080612182616, 27.38311570485433, 27.433946736653645, 27.49376525878906, 27.365997950236004 ], "forward": [ 8.954112021128337, 9.244633642832438, 9.262553596496582, 9.258339246114096, 9.22587521870931 ] }, "usefulTFPerGPU": 177.8648188233958, "nvlinkTXGBps": 25.857254925130448, "nvlinkRXGBps": 25.85230265761001, "nvlinkMaxTXGBps": 114.50335508604464, "nvlinkMaxRXGBps": 114.2012049178209, "analyticalHBMGBps": 125.91209054376458, "receiveRows": [ 98304, 58690, 43404, 35040, 29679, 25913, 23105, 20920, 19164, 17719, 16506, 15471, 14576, 13794, 13104, 12490, 11939, 11442, 10990, 10579, 10202, 9854, 9534, 9237, 8960, 8703, 8462, 8236, 8023, 7823, 7635, 7457, 7288, 7128, 6976, 6831, 6693, 6562, 6436, 6316, 6201, 6091, 5985, 5883, 5786, 5692, 5602, 5514, 5431, 5349, 5271, 5196, 5122, 5052, 4983, 4917, 4853, 4790, 4730, 4671, 4614, 4558, 4504, 4452 ], "slotBytes": 3821010944, "scratchBytes": 5838471168, "arenaBytes": 9659482112, "savedBytes": [ 2793406464, 1738723328, 1331748864, 1109065728, 966334464, 866068480, 791308288, 733134848, 686383104, 647911424, 615616512, 588060672, 564232192, 543412224, 525041664, 508694528, 494024704, 480792576, 468758528, 457816064, 447778816, 438513664, 429993984, 422086656, 414711808, 407869440, 401453056, 395436032, 389765120, 384440320, 379435008, 374695936, 370196480, 365936640, 361889792, 358029312, 354355200, 350867456, 347512832, 344317952, 341256192, 338327552, 335505408, 332789760, 330207232, 327704576, 325308416, 322965504, 320755712, 318572544, 316495872, 314499072, 312528896, 310665216, 308828160, 307070976, 305367040, 303689728, 302092288, 300521472, 299003904, 297512960, 296075264, 294690816 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 11.437751293182373, "reservedPeakGiB": 13.013671875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308224000, "forwardFlopsPerLogicalPayloadByteRank0": 5267.265488372093, "capacityRows": 98304 }, { "id": "training-ep64-mxfp8-mega-single_layer-F3072-i8", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 7.586257028579712, "backwardMS": 10.2489408493042, "layerMS": 17.83841190338135, "windows": { "backward": [ 9.920070362091064, 10.396979236602784, 10.22803093592326, 10.276898193359376, 10.2489408493042 ], "combined": [ 17.13402665456136, 17.97881596883138, 17.785942459106444, 17.884679476420086, 17.83841190338135 ], "forward": [ 7.217447423934937, 7.586257028579712, 7.562117338180542, 7.6119861284891766, 7.593992519378662 ] }, "usefulTFPerGPU": 273.03399764755835, "nvlinkTXGBps": 39.595099759057945, "nvlinkRXGBps": 39.60278458724031, "nvlinkMaxTXGBps": 175.358532569631, "nvlinkMaxRXGBps": 175.6621269815167, "analyticalHBMGBps": 128.70759372726417, "receiveRows": [ 98304, 58690, 43404, 35040, 29679, 25913, 23105, 20920, 19164, 17719, 16506, 15471, 14576, 13794, 13104, 12490, 11939, 11442, 10990, 10579, 10202, 9854, 9534, 9237, 8960, 8703, 8462, 8236, 8023, 7823, 7635, 7457, 7288, 7128, 6976, 6831, 6693, 6562, 6436, 6316, 6201, 6091, 5985, 5883, 5786, 5692, 5602, 5514, 5431, 5349, 5271, 5196, 5122, 5052, 4983, 4917, 4853, 4790, 4730, 4671, 4614, 4558, 4504, 4452 ], "slotBytes": 3566796800, "scratchBytes": 5366722560, "arenaBytes": 8933519360, "savedBytes": [ 2422210560, 1527300096, 1176354816, 983334912, 860504064, 772767744, 720125952, 667484160, 614842368, 597295104, 562200576, 544653312, 509558784, 492011520, 492011520, 474464256, 456916992, 439369728, 439369728, 421822464, 421822464, 404275200, 404275200, 404275200, 386727936, 386727936, 386727936, 369180672, 369180672, 369180672, 351633408, 351633408, 351633408, 351633408, 351633408, 334086144, 334086144, 334086144, 334086144, 334086144, 334086144, 316538880, 316538880, 316538880, 316538880, 316538880, 316538880, 316538880, 316538880, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 281444352, 281444352, 281444352 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 11.153894424438477, "reservedPeakGiB": 12.80078125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308224000, "forwardFlopsPerLogicalPayloadByteRank0": 5267.265488372093, "capacityRows": 98944 }, { "id": "training-ep64-mxfp8-staged-single_layer-F3072-i8", "mode": "training", "ep": 64, "category": "single_layer", "F": 3072, "imbalance": 8.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 6.313576523462931, "backwardMS": 11.507501920064291, "layerMS": 17.818708229064942, "windows": { "backward": [ 11.223789819081624, 11.559975560506185, 11.507501920064291, 11.466866048177083, 11.509015432993571 ], "combined": [ 17.376710573832195, 17.906763776143393, 17.818708229064942, 17.777692794799805, 17.827053769429526 ], "forward": [ 6.155770667394003, 6.350020329157512, 6.313473113377889, 6.313576523462931, 6.320690139134725 ] }, "usefulTFPerGPU": 273.33591476173945, "nvlinkTXGBps": 39.715110567970946, "nvlinkRXGBps": 39.71334362916023, "nvlinkMaxTXGBps": 175.83497107651306, "nvlinkMaxRXGBps": 175.8350848310424, "analyticalHBMGBps": 128.84991675518793, "receiveRows": [ 98304, 58690, 43404, 35040, 29679, 25913, 23105, 20920, 19164, 17719, 16506, 15471, 14576, 13794, 13104, 12490, 11939, 11442, 10990, 10579, 10202, 9854, 9534, 9237, 8960, 8703, 8462, 8236, 8023, 7823, 7635, 7457, 7288, 7128, 6976, 6831, 6693, 6562, 6436, 6316, 6201, 6091, 5985, 5883, 5786, 5692, 5602, 5514, 5431, 5349, 5271, 5196, 5122, 5052, 4983, 4917, 4853, 4790, 4730, 4671, 4614, 4558, 4504, 4452 ], "slotBytes": 3566796800, "scratchBytes": 5024772096, "arenaBytes": 8591568896, "savedBytes": [ 2422210560, 1527300096, 1176354816, 983334912, 860504064, 772767744, 720125952, 667484160, 614842368, 597295104, 562200576, 544653312, 509558784, 492011520, 492011520, 474464256, 456916992, 439369728, 439369728, 421822464, 421822464, 404275200, 404275200, 404275200, 386727936, 386727936, 386727936, 369180672, 369180672, 369180672, 351633408, 351633408, 351633408, 351633408, 351633408, 334086144, 334086144, 334086144, 334086144, 334086144, 334086144, 316538880, 316538880, 316538880, 316538880, 316538880, 316538880, 316538880, 316538880, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 298991616, 281444352, 281444352, 281444352 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 10.835428237915039, "reservedPeakGiB": 12.48828125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 308224000, "forwardFlopsPerLogicalPayloadByteRank0": 5267.265488372093, "capacityRows": 98944 }, { "id": "training-ep64-bf16-standard-width-F1536-i1", "mode": "training", "ep": 64, "category": "width", "F": 1536, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 0.9089269359906514, "backwardMS": 1.456173872947693, "layerMS": 2.363606397310893, "windows": { "backward": [ 1.4295168002446492, 1.456173872947693, 1.4616479992866516, 1.4685183882713317, 1.4410943865776062 ], "combined": [ 2.322054394086202, 2.363606397310893, 2.3703157186508177, 2.378986692428589, 2.34136106967926 ], "forward": [ 0.8939317365487417, 0.9089269359906514, 0.9099125385284423, 0.9124746680259704, 0.9021333316961925 ] }, "usefulTFPerGPU": 1030.3096402186984, "nvlinkTXGBps": 294.4624901908178, "nvlinkRXGBps": 294.4632552503558, "nvlinkMaxTXGBps": 299.76395495388874, "nvlinkMaxRXGBps": 299.7648228363287, "analyticalHBMGBps": 859.8249464513895, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3217031168, "scratchBytes": 4328521728, "arenaBytes": 7545552896, "savedBytes": [ 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008, 427819008 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 8.361573696136475, "reservedPeakGiB": 9.185546875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 2340.7649441918147, "capacityRows": 98304 }, { "id": "training-ep64-mxfp8-mega-width-F1536-i1", "mode": "training", "ep": 64, "category": "width", "F": 1536, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 0.8318709353605906, "backwardMS": 0.9632117370764415, "layerMS": 1.7940927982330321, "windows": { "backward": [ 0.9633525331815084, 0.9625536024570465, 0.9629312018553416, 0.9632117370764415, 0.9637194693088531 ], "combined": [ 1.796996259689331, 1.7925077279408772, 1.7940927982330321, 1.79187521537145, 1.7951242566108703 ], "forward": [ 0.834087469180425, 0.831118929386139, 0.8318709353605906, 0.8297130644321442, 0.8319285333156585 ] }, "usefulTFPerGPU": 1357.3692839246821, "nvlinkTXGBps": 385.18092393171787, "nvlinkRXGBps": 385.1819604802766, "nvlinkMaxTXGBps": 396.3449660634971, "nvlinkMaxRXGBps": 396.3450106119817, "analyticalHBMGBps": 811.7331552940329, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 2767536128, "scratchBytes": 3813697536, "arenaBytes": 6581233664, "savedBytes": [ 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 7.659624099731445, "reservedPeakGiB": 8.501953125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 2340.7649441918147, "capacityRows": 98944 }, { "id": "training-ep64-mxfp8-staged-width-F1536-i1", "mode": "training", "ep": 64, "category": "width", "F": 1536, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 0.8371882617473603, "backwardMS": 1.1280757308006286, "layerMS": 1.9645312070846557, "windows": { "backward": [ 1.1284714659055075, 1.1277994712193806, 1.1276512066523234, 1.1286762674649558, 1.1280757308006286 ], "combined": [ 1.965992530186971, 1.965062403678894, 1.96310080687205, 1.9645312070846557, 1.9644159952799478 ], "forward": [ 0.8382751981417338, 0.8380192041397094, 0.8361717323462169, 0.8362730741500854, 0.8371882617473603 ] }, "usefulTFPerGPU": 1239.6069087881178, "nvlinkTXGBps": 354.33503986864935, "nvlinkRXGBps": 354.3355995012589, "nvlinkMaxTXGBps": 362.6631796965071, "nvlinkMaxRXGBps": 362.66334583980614, "analyticalHBMGBps": 741.3089711927615, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 2767536128, "scratchBytes": 3642722304, "arenaBytes": 6410258432, "savedBytes": [ 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328, 361955328 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 7.500391006469727, "reservedPeakGiB": 8.345703125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 2340.7649441918147, "capacityRows": 98944 }, { "id": "training-ep64-bf16-standard-width-F3072-i1", "mode": "training", "ep": 64, "category": "width", "F": 3072, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 1.36560746828715, "backwardMS": 2.4566965341567992, "layerMS": 3.820039470990499, "windows": { "backward": [ 2.377964814503988, 2.4566965341567992, 2.4552053213119507, 2.4997813224792482, 2.470046941439311 ], "combined": [ 3.703446388244629, 3.820039470990499, 3.8191829284032184, 3.8776554505030316, 3.835208519299825 ], "forward": [ 1.327181871732076, 1.36560746828715, 1.3652565280596416, 1.3801002701123555, 1.3673184076944986 ] }, "usefulTFPerGPU": 1274.9849708754784, "nvlinkTXGBps": 175.11838930237465, "nvlinkRXGBps": 175.11848944603378, "nvlinkMaxTXGBps": 176.34056916084475, "nvlinkMaxRXGBps": 176.34048674576232, "analyticalHBMGBps": 902.5732247489061, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3821010944, "scratchBytes": 5838471168, "arenaBytes": 9659482112, "savedBytes": [ 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480, 503316480 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 11.437754154205322, "reservedPeakGiB": 13.013671875, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 4681.529888383629, "capacityRows": 98304 }, { "id": "training-ep64-mxfp8-mega-width-F3072-i1", "mode": "training", "ep": 64, "category": "width", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 1.155949858824412, "backwardMS": 1.4970933357874552, "layerMS": 2.6506304264068605, "windows": { "backward": [ 1.4822239995002746, 1.4970933357874552, 1.5126229365666708, 1.5043978532155355, 1.4909493366877238 ], "combined": [ 2.575806903839111, 2.6527093331019085, 2.679684281349182, 2.6506304264068605, 2.645743981997172 ], "forward": [ 1.0945365309715271, 1.1562751928965251, 1.1678944071133932, 1.147488001982371, 1.155949858824412 ] }, "usefulTFPerGPU": 1837.484722555735, "nvlinkTXGBps": 260.8373412167566, "nvlinkRXGBps": 260.83797680137826, "nvlinkMaxTXGBps": 264.13220603156935, "nvlinkMaxRXGBps": 264.13330247266396, "analyticalHBMGBps": 866.1860397914195, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3566796800, "scratchBytes": 5366722560, "arenaBytes": 8933519360, "savedBytes": [ 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 11.153897285461426, "reservedPeakGiB": 12.80078125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 4681.529888383629, "capacityRows": 98944 }, { "id": "training-ep64-mxfp8-staged-width-F3072-i1", "mode": "training", "ep": 64, "category": "width", "F": 3072, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 0.9711893300215403, "backwardMS": 1.634441598256429, "layerMS": 2.6033589204152423, "windows": { "backward": [ 1.6001898646354675, 1.6352991938591004, 1.6407050609588623, 1.6318965355555217, 1.634441598256429 ], "combined": [ 2.557537070910136, 2.6033589204152423, 2.6118794679641724, 2.601818648974101, 2.60545388062795 ], "forward": [ 0.9588554680347443, 0.9689333359400432, 0.9722389340400696, 0.9711893300215403, 0.9724010646343231 ] }, "usefulTFPerGPU": 1870.8495687898248, "nvlinkTXGBps": 264.98620956176387, "nvlinkRXGBps": 264.9869158388584, "nvlinkMaxTXGBps": 268.7157346819157, "nvlinkMaxRXGBps": 268.7239652359081, "analyticalHBMGBps": 881.914150982221, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 3566796800, "scratchBytes": 5024772096, "arenaBytes": 8591568896, "savedBytes": [ 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992, 456916992 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 10.835431098937988, "reservedPeakGiB": 12.48828125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 4681.529888383629, "capacityRows": 98944 }, { "id": "training-ep64-bf16-standard-width-F6144-i1", "mode": "training", "ep": 64, "category": "width", "F": 6144, "imbalance": 1.0, "precision": "bf16", "pipeline": "standard", "forwardMS": 2.54978879292806, "backwardMS": 4.971671454111735, "layerMS": 7.519784545898437, "windows": { "backward": [ 4.758985630671183, 4.904314692815145, 4.971671454111735, 4.977554178237915, 5.0083925247192385 ], "combined": [ 7.208100239435832, 7.41572691599528, 7.519784545898437, 7.529861323038737, 7.566156800587972 ], "forward": [ 2.450702929496765, 2.512827730178833, 2.54978879292806, 2.554094942410787, 2.5592170635859173 ] }, "usefulTFPerGPU": 1295.3809737329625, "nvlinkTXGBps": 93.45099867016964, "nvlinkRXGBps": 93.451078321407, "nvlinkMaxTXGBps": 93.63246709157264, "nvlinkMaxRXGBps": 93.63192742929648, "analyticalHBMGBps": 835.000059599421, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 5028970496, "scratchBytes": 8858370048, "arenaBytes": 13887340544, "savedBytes": [ 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424, 654311424 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 17.590089321136475, "reservedPeakGiB": 20.650390625, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 9363.059776767259, "capacityRows": 98304 }, { "id": "training-ep64-mxfp8-mega-width-F6144-i1", "mode": "training", "ep": 64, "category": "width", "F": 6144, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "mega", "forwardMS": 1.9457119981447855, "backwardMS": 2.7785120169321695, "layerMS": 4.723643732070923, "windows": { "backward": [ 2.761050653457642, 2.7771680037180584, 2.783892265955607, 2.7785120169321695, 2.861962668100993 ], "combined": [ 4.656744575500488, 4.717736514409383, 4.728690163294474, 4.723643732070923, 4.849207480748494 ], "forward": [ 1.8962176044782002, 1.9416074633598328, 1.9457119981447855, 1.9460000077883401, 1.9882410685221354 ] }, "usefulTFPerGPU": 2062.17623086011, "nvlinkTXGBps": 145.69456781973946, "nvlinkRXGBps": 145.69437509711423, "nvlinkMaxTXGBps": 152.8902032282228, "nvlinkMaxRXGBps": 152.8893122846551, "analyticalHBMGBps": 841.5469551631959, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 5165318144, "scratchBytes": 8472772608, "arenaBytes": 13638090752, "savedBytes": [ 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 18.142417907714844, "reservedPeakGiB": 21.412109375, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 9363.059776767259, "capacityRows": 98944 }, { "id": "training-ep64-mxfp8-staged-width-F6144-i1", "mode": "training", "ep": 64, "category": "width", "F": 6144, "imbalance": 1.0, "precision": "mxfp8", "pipeline": "staged", "forwardMS": 1.659659731388092, "backwardMS": 2.9031914552052815, "layerMS": 4.559766435623169, "windows": { "backward": [ 2.817318399747213, 2.9031914552052815, 2.9012128114700317, 2.9282442728678384, 2.976475739479065 ], "combined": [ 4.4310357411702475, 4.55970238049825, 4.559766435623169, 4.593933868408203, 4.664040549596151 ], "forward": [ 1.6153280059496562, 1.658136534690857, 1.659659731388092, 1.6675136049588521, 1.6896725416183471 ] }, "usefulTFPerGPU": 2136.2905238361686, "nvlinkTXGBps": 150.91287867643476, "nvlinkRXGBps": 150.91303797038103, "nvlinkMaxTXGBps": 157.46162630360513, "nvlinkMaxRXGBps": 157.4625735527007, "analyticalHBMGBps": 871.7920218334003, "receiveRows": [ 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288, 12288 ], "slotBytes": 5165318144, "scratchBytes": 8358789120, "arenaBytes": 13524107264, "savedBytes": [ 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320, 646840320 ], "inputBytes": 29360128, "hostScratchBytes": 0, "recompute": [ false ], "allocatedPeakGiB": 18.03626251220703, "reservedPeakGiB": 21.314453125, "numericalAccepted": true, "graphAccepted": true, "allocationRetries": 0, "ooms": 0, "logicalForwardPayloadBytesRank0": 346787840, "forwardFlopsPerLogicalPayloadByteRank0": 9363.059776767259, "capacityRows": 98944 } ] }