Budget3D-Idol / perf /optimised.json
ozgurkara's picture
Budget3D-Idol: release
a689103
Raw History Blame Contribute Delete
23.4 kB
{ "gallery": "Budget3D-Idol, optimised inference", "objects": 98, "objects_with_perf": 98, "sampling_calls": 63, "dropped_warmup_calls": 21, "dropped_other_batch_calls": 14, "gpus": [ "NVIDIA H100 80GB HBM3" ], "turbo": false, "precision": "bf16 transformer blocks, TF32 elsewhere, cuDNN fused attention, torch.compile (blocks + adapters)", "steps": 50, "cfg": 3.0, "num_gaussians": 262144, "frames": 72, "size": 384, "per_budget": [ { "budget": 1.0, "tokens": 8192, "n4": 0, "n2": 0, "n1": 8192, "batches": 9, "batch_sizes": [ 8 ], "sample_s_per_obj": { "n": 9, "mean": 5.667208333333334, "std": 0.02751292309988177, "median": 5.673375, "p10": 5.642125, "p90": 5.6885, "min": 5.621375, "max": 5.711 }, "ms_per_step_per_obj": 113.34416666666668, "throughput_obj_per_min": 10.587223280127635, "speedup_paired": { "n": 9, "mean": 1.0, "std": 0.0, "median": 1.0, "p10": 1.0, "p90": 1.0, "min": 1.0, "max": 1.0 }, "speedup_of_means": 1.0, "token_ratio": 1.0, "sample_peak_alloc_gb": { "n": 9, "mean": 9.90966796875, "std": 0.0, "median": 9.90966796875, "p10": 9.90966796875, "p90": 9.90966796875, "min": 9.90966796875, "max": 9.90966796875 }, "sample_peak_reserved_gb": { "n": 9, "mean": 13.823567708333334, "std": 0.001953125, "median": 13.822265625, "p10": 13.822265625, "p90": 13.826171875, "min": 13.822265625, "max": 13.826171875 }, "sample_avg_alloc_gb": { "n": 9, "mean": 4.825455729166666, "std": 0.005957831657292742, "median": 4.82744140625, "p10": 4.8201171875, "p90": 4.82880859375, "min": 4.81474609375, "max": 4.83447265625 }, "sample_base_alloc_gb": { "n": 9, "mean": 4.50732421875, "std": 0.0, "median": 4.50732421875, "p10": 4.50732421875, "p90": 4.50732421875, "min": 4.50732421875, "max": 4.50732421875 }, "sample_peak_minus_base_per_obj_gb": { "n": 9, "mean": 0.67529296875, "std": 0.0, "median": 0.67529296875, "p10": 0.67529296875, "p90": 0.67529296875, "min": 0.67529296875, "max": 0.67529296875 }, "decode_s": { "n": 72, "mean": 2.5310277777777777, "std": 0.7288177378413448, "median": 2.624, "p10": 1.65, "p90": 3.142, "min": 0.62, "max": 6.219 }, "decode_peak_alloc_gb": { "n": 72, "mean": 4.71494140625, "std": 0.0, "median": 4.71494140625, "p10": 4.71494140625, "p90": 4.71494140625, "min": 4.71494140625, "max": 4.71494140625 }, "render_s_per_orbit": { "n": 72, "mean": 3.5987222222222224, "std": 0.8920027825103797, "median": 3.503, "p10": 2.587, "p90": 4.783, "min": 2.086, "max": 7.146 }, "render_fps": { "n": 72, "mean": 21.13486111111111, "std": 4.858344351689484, "median": 20.555, "p10": 15.05, "p90": 27.83, "min": 10.08, "max": 34.51 }, "render_peak_alloc_gb": { "n": 72, "mean": 4.963486056857639, "std": 0.09088779811047575, "median": 4.947900390625, "p10": 4.85244140625, "p90": 5.11064453125, "min": 4.82431640625, "max": 5.1625 }, "end_to_end_s_per_obj": { "n": 72, "mean": 11.796958333333334, "std": 1.114830620937675, "median": 11.6001875, "p10": 10.647875, "p90": 13.353375, "min": 9.231875, "max": 15.024375000000001 } }, { "budget": 0.9, "tokens": 7373, "n4": 21, "n2": 504, "n1": 6848, "batches": 9, "batch_sizes": [ 8 ], "sample_s_per_obj": { "n": 9, "mean": 5.149319444444444, "std": 0.3415163916220242, "median": 5.01675, "p10": 4.989625, "p90": 5.23075, "min": 4.987625, "max": 6.037625 }, "ms_per_step_per_obj": 102.98638888888888, "throughput_obj_per_min": 11.652025213687894, "speedup_paired": { "n": 9, "mean": 1.1042555213698546, "std": 0.06262438488800678, "median": 1.1276569173514501, "p10": 1.0853128136500503, "p90": 1.1335003845485896, "min": 0.9421751102461647, "max": 1.135330811433725 }, "speedup_of_means": 1.1005742398537026, "token_ratio": 1.1110809711108096, "sample_peak_alloc_gb": { "n": 9, "mean": 9.56083984375, "std": 0.0, "median": 9.56083984375, "p10": 9.56083984375, "p90": 9.56083984375, "min": 9.56083984375, "max": 9.56083984375 }, "sample_peak_reserved_gb": { "n": 9, "mean": 13.823567708333334, "std": 0.001953125, "median": 13.822265625, "p10": 13.822265625, "p90": 13.826171875, "min": 13.822265625, "max": 13.826171875 }, "sample_avg_alloc_gb": { "n": 9, "mean": 6.26328125, "std": 0.00864630150197672, "median": 6.26259765625, "p10": 6.25146484375, "p90": 6.27080078125, "min": 6.2505859375, "max": 6.2763671875 }, "sample_base_alloc_gb": { "n": 9, "mean": 4.5076171875, "std": 0.0, "median": 4.5076171875, "p10": 4.5076171875, "p90": 4.5076171875, "min": 4.5076171875, "max": 4.5076171875 }, "sample_peak_minus_base_per_obj_gb": { "n": 9, "mean": 0.6316528320312499, "std": 0.0, "median": 0.6316528320312499, "p10": 0.6316528320312499, "p90": 0.6316528320312499, "min": 0.6316528320312499, "max": 0.6316528320312499 }, "decode_s": { "n": 72, "mean": 2.1547916666666667, "std": 0.7236627674934907, "median": 2.111, "p10": 1.114, "p90": 2.852, "min": 0.108, "max": 4.438 }, "decode_peak_alloc_gb": { "n": 72, "mean": 4.71494140625, "std": 0.0, "median": 4.71494140625, "p10": 4.71494140625, "p90": 4.71494140625, "min": 4.71494140625, "max": 4.71494140625 }, "render_s_per_orbit": { "n": 72, "mean": 3.542472222222222, "std": 0.8503987129861577, "median": 3.3834999999999997, "p10": 2.606, "p90": 4.721, "min": 2.235, "max": 6.838 }, "render_fps": { "n": 72, "mean": 21.39875, "std": 4.736149841052478, "median": 21.28, "p10": 15.25, "p90": 27.63, "min": 10.53, "max": 32.21 }, "render_peak_alloc_gb": { "n": 72, "mean": 4.965403917100694, "std": 0.09349110849140592, "median": 4.954443359375, "p10": 4.8447265625, "p90": 5.1171875, "min": 4.8279296875, "max": 5.16796875 }, "end_to_end_s_per_obj": { "n": 72, "mean": 10.846583333333333, "std": 1.2017942403451436, "median": 10.680625000000001, "p10": 9.384625, "p90": 12.14375, "min": 8.961625, "max": 14.438749999999999 } }, { "budget": 0.8, "tokens": 6554, "n4": 69, "n2": 603, "n1": 5882, "batches": 9, "batch_sizes": [ 8 ], "sample_s_per_obj": { "n": 9, "mean": 4.589138888888889, "std": 0.25731087141551384, "median": 4.504375, "p10": 4.484, "p90": 4.524, "min": 4.479625, "max": 5.274 }, "ms_per_step_per_obj": 91.78277777777777, "throughput_obj_per_min": 13.074348249792687, "speedup_paired": { "n": 9, "mean": 1.2378896277148217, "std": 0.05986946048766982, "median": 1.2548861519641914, "p10": 1.2536518733273863, "p90": 1.2633907585100796, "min": 1.0785930982176717, "max": 1.2645868794820994 }, "speedup_of_means": 1.2349175892354534, "token_ratio": 1.2499237107110162, "sample_peak_alloc_gb": { "n": 9, "mean": 9.21005859375, "std": 0.0, "median": 9.21005859375, "p10": 9.21005859375, "p90": 9.21005859375, "min": 9.21005859375, "max": 9.21005859375 }, "sample_peak_reserved_gb": { "n": 9, "mean": 13.823567708333334, "std": 0.001953125, "median": 13.822265625, "p10": 13.822265625, "p90": 13.826171875, "min": 13.822265625, "max": 13.826171875 }, "sample_avg_alloc_gb": { "n": 9, "mean": 6.1707573784722225, "std": 0.006234170152384163, "median": 6.17099609375, "p10": 6.1640625, "p90": 6.17783203125, "min": 6.1638671875, "max": 6.18017578125 }, "sample_base_alloc_gb": { "n": 9, "mean": 4.5076171875, "std": 0.0, "median": 4.5076171875, "p10": 4.5076171875, "p90": 4.5076171875, "min": 4.5076171875, "max": 4.5076171875 }, "sample_peak_minus_base_per_obj_gb": { "n": 9, "mean": 0.58780517578125, "std": 0.0, "median": 0.58780517578125, "p10": 0.58780517578125, "p90": 0.58780517578125, "min": 0.58780517578125, "max": 0.58780517578125 }, "decode_s": { "n": 72, "mean": 2.1701111111111113, "std": 0.504645923997409, "median": 2.101, "p10": 1.6, "p90": 2.626, "min": 1.1, "max": 4.547 }, "decode_peak_alloc_gb": { "n": 72, "mean": 4.71494140625, "std": 0.0, "median": 4.71494140625, "p10": 4.71494140625, "p90": 4.71494140625, "min": 4.71494140625, "max": 4.71494140625 }, "render_s_per_orbit": { "n": 72, "mean": 3.582152777777778, "std": 1.0026080871059257, "median": 3.399, "p10": 2.531, "p90": 4.624, "min": 2.078, "max": 7.323 }, "render_fps": { "n": 72, "mean": 21.441388888888888, "std": 5.182956985567312, "median": 21.185000000000002, "p10": 15.57, "p90": 28.45, "min": 9.83, "max": 34.66 }, "render_peak_alloc_gb": { "n": 72, "mean": 4.967967393663194, "std": 0.09382631946413707, "median": 4.961865234375, "p10": 4.8513671875, "p90": 5.1083984375, "min": 4.8275390625, "max": 5.1787109375 }, "end_to_end_s_per_obj": { "n": 72, "mean": 10.341402777777779, "std": 1.1827926182296937, "median": 10.2583125, "p10": 9.141, "p90": 11.801, "min": 8.448, "max": 14.453375 } }, { "budget": 0.7, "tokens": 5734, "n4": 127, "n2": 553, "n1": 5054, "batches": 9, "batch_sizes": [ 8 ], "sample_s_per_obj": { "n": 9, "mean": 4.098944444444444, "std": 0.2863670619747108, "median": 3.99925, "p10": 3.973, "p90": 4.10925, "min": 3.963, "max": 4.85425 }, "ms_per_step_per_obj": 81.97888888888889, "throughput_obj_per_min": 14.637914910342772, "speedup_paired": { "n": 9, "mean": 1.3879020467897893, "std": 0.08553096732974155, "median": 1.4184645470603077, "p10": 1.3843158727261666, "p90": 1.4253891991389263, "min": 1.1623062265025492, "max": 1.4258431915429148 }, "speedup_of_means": 1.3826018893753136, "token_ratio": 1.4286710847575863, "sample_peak_alloc_gb": { "n": 9, "mean": 8.85966796875, "std": 0.0, "median": 8.85966796875, "p10": 8.85966796875, "p90": 8.85966796875, "min": 8.85966796875, "max": 8.85966796875 }, "sample_peak_reserved_gb": { "n": 9, "mean": 13.823567708333334, "std": 0.001953125, "median": 13.822265625, "p10": 13.822265625, "p90": 13.826171875, "min": 13.822265625, "max": 13.826171875 }, "sample_avg_alloc_gb": { "n": 9, "mean": 6.06259765625, "std": 0.0060668822505220545, "median": 6.06298828125, "p10": 6.05458984375, "p90": 6.06875, "min": 6.05419921875, "max": 6.07275390625 }, "sample_base_alloc_gb": { "n": 9, "mean": 4.50751953125, "std": 0.0, "median": 4.50751953125, "p10": 4.50751953125, "p90": 4.50751953125, "min": 4.50751953125, "max": 4.50751953125 }, "sample_peak_minus_base_per_obj_gb": { "n": 9, "mean": 0.5440185546874999, "std": 0.0, "median": 0.5440185546874999, "p10": 0.5440185546874999, "p90": 0.5440185546874999, "min": 0.5440185546874999, "max": 0.5440185546874999 }, "decode_s": { "n": 72, "mean": 2.0640972222222222, "std": 0.4807115371654879, "median": 2.1020000000000003, "p10": 1.595, "p90": 2.662, "min": 0.606, "max": 3.122 }, "decode_peak_alloc_gb": { "n": 72, "mean": 4.71484375, "std": 0.0, "median": 4.71484375, "p10": 4.71484375, "p90": 4.71484375, "min": 4.71484375, "max": 4.71484375 }, "render_s_per_orbit": { "n": 72, "mean": 3.4767638888888888, "std": 0.8333036998252258, "median": 3.318, "p10": 2.538, "p90": 4.539, "min": 2.04, "max": 6.604 }, "render_fps": { "n": 72, "mean": 21.83361111111111, "std": 4.967508885013866, "median": 21.7, "p10": 15.86, "p90": 28.37, "min": 10.9, "max": 35.3 }, "render_peak_alloc_gb": { "n": 72, "mean": 4.973456488715278, "std": 0.0957374973066442, "median": 4.97666015625, "p10": 4.84921875, "p90": 5.10791015625, "min": 4.823046875, "max": 5.183203125 }, "end_to_end_s_per_obj": { "n": 72, "mean": 9.639805555555556, "std": 1.0072361152880354, "median": 9.494625, "p10": 8.51925, "p90": 10.813, "min": 7.843624999999999, "max": 13.154375000000002 } }, { "budget": 0.6, "tokens": 4915, "n4": 186, "n2": 487, "n1": 4242, "batches": 9, "batch_sizes": [ 8 ], "sample_s_per_obj": { "n": 9, "mean": 3.6000277777777776, "std": 0.2599901348622568, "median": 3.510375, "p10": 3.475875, "p90": 3.598375, "min": 3.475375, "max": 4.286125 }, "ms_per_step_per_obj": 72.00055555555555, "throughput_obj_per_min": 16.666538066835905, "speedup_paired": { "n": 9, "mean": 1.5806288997508535, "std": 0.10015510486874488, "median": 1.6174873215120669, "p10": 1.5808524681279745, "p90": 1.6239802490339201, "min": 1.3163696812388812, "max": 1.6297694825044056 }, "speedup_of_means": 1.5742123903364946, "token_ratio": 1.666734486266531, "sample_peak_alloc_gb": { "n": 9, "mean": 8.5111328125, "std": 0.0, "median": 8.5111328125, "p10": 8.5111328125, "p90": 8.5111328125, "min": 8.5111328125, "max": 8.5111328125 }, "sample_peak_reserved_gb": { "n": 9, "mean": 13.823567708333334, "std": 0.001953125, "median": 13.822265625, "p10": 13.822265625, "p90": 13.826171875, "min": 13.822265625, "max": 13.826171875 }, "sample_avg_alloc_gb": { "n": 9, "mean": 5.9558702256944445, "std": 0.004790246725468664, "median": 5.95634765625, "p10": 5.9494140625, "p90": 5.959375, "min": 5.94931640625, "max": 5.96435546875 }, "sample_base_alloc_gb": { "n": 9, "mean": 4.50751953125, "std": 0.0, "median": 4.50751953125, "p10": 4.50751953125, "p90": 4.50751953125, "min": 4.50751953125, "max": 4.50751953125 }, "sample_peak_minus_base_per_obj_gb": { "n": 9, "mean": 0.50045166015625, "std": 0.0, "median": 0.50045166015625, "p10": 0.50045166015625, "p90": 0.50045166015625, "min": 0.50045166015625, "max": 0.50045166015625 }, "decode_s": { "n": 72, "mean": 2.1553333333333335, "std": 0.5309688991183814, "median": 2.161, "p10": 1.591, "p90": 2.683, "min": 0.628, "max": 3.27 }, "decode_peak_alloc_gb": { "n": 72, "mean": 4.71484375, "std": 0.0, "median": 4.71484375, "p10": 4.71484375, "p90": 4.71484375, "min": 4.71484375, "max": 4.71484375 }, "render_s_per_orbit": { "n": 72, "mean": 3.5778194444444447, "std": 0.879133685717404, "median": 3.3505000000000003, "p10": 2.575, "p90": 4.674, "min": 2.335, "max": 6.625 }, "render_fps": { "n": 72, "mean": 21.281805555555554, "std": 4.953212736433099, "median": 21.485, "p10": 15.4, "p90": 27.96, "min": 10.87, "max": 30.84 }, "render_peak_alloc_gb": { "n": 72, "mean": 4.979347059461806, "std": 0.09405926622217514, "median": 4.974609375, "p10": 4.8623046875, "p90": 5.10400390625, "min": 4.8322265625, "max": 5.194140625 }, "end_to_end_s_per_obj": { "n": 72, "mean": 9.333180555555556, "std": 1.0915755243643184, "median": 9.3613125, "p10": 8.029375, "p90": 10.729125, "min": 7.189874999999999, "max": 12.6875 } }, { "budget": 0.5, "tokens": 4096, "n4": 245, "n2": 421, "n1": 3430, "batches": 9, "batch_sizes": [ 8 ], "sample_s_per_obj": { "n": 9, "mean": 3.1818333333333335, "std": 0.31655081419694686, "median": 3.067875, "p10": 3.05375, "p90": 3.163, "min": 3.0455, "max": 4.021 }, "ms_per_step_per_obj": 63.63666666666667, "throughput_obj_per_min": 18.857053061652085, "speedup_paired": { "n": 9, "mean": 1.7941564644004022, "std": 0.14765573964574605, "median": 1.849524041981938, "p10": 1.7984508378122037, "p90": 1.8526818515797208, "min": 1.4031646356627705, "max": 1.8550552599263201 }, "speedup_of_means": 1.781114137551726, "token_ratio": 2.0, "sample_peak_alloc_gb": { "n": 9, "mean": 8.1595703125, "std": 0.0, "median": 8.1595703125, "p10": 8.1595703125, "p90": 8.1595703125, "min": 8.1595703125, "max": 8.1595703125 }, "sample_peak_reserved_gb": { "n": 9, "mean": 13.823567708333334, "std": 0.001953125, "median": 13.822265625, "p10": 13.822265625, "p90": 13.826171875, "min": 13.822265625, "max": 13.826171875 }, "sample_avg_alloc_gb": { "n": 9, "mean": 5.849240451388889, "std": 0.005109199585724242, "median": 5.84794921875, "p10": 5.84560546875, "p90": 5.851953125, "min": 5.84228515625, "max": 5.85986328125 }, "sample_base_alloc_gb": { "n": 9, "mean": 4.507421875, "std": 0.0, "median": 4.507421875, "p10": 4.507421875, "p90": 4.507421875, "min": 4.507421875, "max": 4.507421875 }, "sample_peak_minus_base_per_obj_gb": { "n": 9, "mean": 0.4565185546874999, "std": 0.0, "median": 0.4565185546874999, "p10": 0.4565185546874999, "p90": 0.4565185546874999, "min": 0.4565185546874999, "max": 0.4565185546874999 }, "decode_s": { "n": 72, "mean": 2.137583333333333, "std": 0.44352082854372543, "median": 2.159, "p10": 1.607, "p90": 2.623, "min": 0.615, "max": 3.124 }, "decode_peak_alloc_gb": { "n": 72, "mean": 4.71474609375, "std": 0.0, "median": 4.71474609375, "p10": 4.71474609375, "p90": 4.71474609375, "min": 4.71474609375, "max": 4.71474609375 }, "render_s_per_orbit": { "n": 72, "mean": 3.636763888888889, "std": 0.8945983833039358, "median": 3.517, "p10": 2.631, "p90": 4.664, "min": 2.114, "max": 6.479 }, "render_fps": { "n": 72, "mean": 20.91402777777778, "std": 4.8100796923860685, "median": 20.47, "p10": 15.44, "p90": 27.37, "min": 11.11, "max": 34.06 }, "render_peak_alloc_gb": { "n": 72, "mean": 4.992020670572916, "std": 0.09525087901734988, "median": 4.98974609375, "p10": 4.87177734375, "p90": 5.10859375, "min": 4.83935546875, "max": 5.2017578125 }, "end_to_end_s_per_obj": { "n": 72, "mean": 8.956180555555555, "std": 1.0426814018063841, "median": 8.804375, "p10": 7.642750000000001, "p90": 10.046375000000001, "min": 6.56875, "max": 11.65025 } }, { "budget": 0.4, "tokens": 3277, "n4": 305, "n2": 340, "n1": 2632, "batches": 9, "batch_sizes": [ 8 ], "sample_s_per_obj": { "n": 9, "mean": 2.6642916666666667, "std": 0.24473693625758652, "median": 2.583, "p10": 2.5685, "p90": 2.6235, "min": 2.56675, "max": 3.31525 }, "ms_per_step_per_obj": 53.28583333333333, "throughput_obj_per_min": 22.52005692569945, "speedup_paired": { "n": 9, "mean": 2.140546197915536, "std": 0.16492962624862212, "median": 2.1899893914552995, "p10": 2.1768629693157995, "p90": 2.204982241035372, "min": 1.7018701455395522, "max": 2.216226745884874 }, "speedup_of_means": 2.1270975712744162, "token_ratio": 2.4998474214220323, "sample_peak_alloc_gb": { "n": 9, "mean": 7.8095703125, "std": 0.0, "median": 7.8095703125, "p10": 7.8095703125, "p90": 7.8095703125, "min": 7.8095703125, "max": 7.8095703125 }, "sample_peak_reserved_gb": { "n": 9, "mean": 13.823567708333334, "std": 0.001953125, "median": 13.822265625, "p10": 13.822265625, "p90": 13.826171875, "min": 13.822265625, "max": 13.826171875 }, "sample_avg_alloc_gb": { "n": 9, "mean": 5.743109809027778, "std": 0.006827755830524479, "median": 5.74326171875, "p10": 5.73583984375, "p90": 5.75205078125, "min": 5.7341796875, "max": 5.7521484375 }, "sample_base_alloc_gb": { "n": 9, "mean": 4.507421875, "std": 0.0, "median": 4.507421875, "p10": 4.507421875, "p90": 4.507421875, "min": 4.507421875, "max": 4.507421875 }, "sample_peak_minus_base_per_obj_gb": { "n": 9, "mean": 0.41276855468749996, "std": 0.0, "median": 0.41276855468749996, "p10": 0.41276855468749996, "p90": 0.41276855468749996, "min": 0.41276855468749996, "max": 0.41276855468749996 }, "decode_s": { "n": 72, "mean": 2.164388888888889, "std": 0.4781613421958942, "median": 2.129, "p10": 1.61, "p90": 2.707, "min": 1.117, "max": 3.206 }, "decode_peak_alloc_gb": { "n": 72, "mean": 4.71474609375, "std": 0.0, "median": 4.71474609375, "p10": 4.71474609375, "p90": 4.71474609375, "min": 4.71474609375, "max": 4.71474609375 }, "render_s_per_orbit": { "n": 72, "mean": 3.795166666666667, "std": 0.8607757065004846, "median": 3.635, "p10": 2.768, "p90": 5.017, "min": 2.334, "max": 6.462 }, "render_fps": { "n": 72, "mean": 19.92222222222222, "std": 4.391115942164063, "median": 19.805, "p10": 14.35, "p90": 26.01, "min": 11.14, "max": 30.85 }, "render_peak_alloc_gb": { "n": 72, "mean": 5.003263346354166, "std": 0.09188556377589595, "median": 5.011865234375, "p10": 4.88017578125, "p90": 5.116015625, "min": 4.83935546875, "max": 5.21279296875 }, "end_to_end_s_per_obj": { "n": 72, "mean": 8.623847222222222, "std": 0.9845748989345064, "median": 8.634, "p10": 7.33425, "p90": 9.9545, "min": 6.55225, "max": 11.60825 } } ], "nvidia_smi": { "0": { "samples": 895, "util_mean": 33.02905027932961, "mem_mean_mib": 13113.467039106145, "mem_peak_mib": 14973.0, "power_mean_w": 265.2586815642458, "power_peak_w": 701.2 }, "1": { "samples": 895, "util_mean": 32.23463687150838, "mem_mean_mib": 13179.66592178771, "mem_peak_mib": 14971.0, "power_mean_w": 260.88100558659215, "power_peak_w": 702.22 }, "2": { "samples": 895, "util_mean": 31.953072625698326, "mem_mean_mib": 12727.979888268157, "mem_peak_mib": 14963.0, "power_mean_w": 249.60315083798884, "power_peak_w": 707.1 } } }
{
"gallery": "Budget3D-Idol, optimised inference",
"objects": 98,
"objects_with_perf": 98,
"sampling_calls": 63,
"dropped_warmup_calls": 21,
"dropped_other_batch_calls": 14,
"gpus": [
"NVIDIA H100 80GB HBM3"
],
"turbo": false,
"precision": "bf16 transformer blocks, TF32 elsewhere, cuDNN fused attention, torch.compile (blocks + adapters)",
"steps": 50,
"cfg": 3.0,
"num_gaussians": 262144,
"frames": 72,
"size": 384,
"per_budget": [
{
"budget": 1.0,
"tokens": 8192,
"n4": 0,
"n2": 0,
"n1": 8192,
"batches": 9,
"batch_sizes": [
8
],
"sample_s_per_obj": {
"n": 9,
"mean": 5.667208333333334,
"std": 0.02751292309988177,
"median": 5.673375,
"p10": 5.642125,
"p90": 5.6885,
"min": 5.621375,
"max": 5.711
},
"ms_per_step_per_obj": 113.34416666666668,
"throughput_obj_per_min": 10.587223280127635,
"speedup_paired": {
"n": 9,
"mean": 1.0,
"std": 0.0,
"median": 1.0,
"p10": 1.0,
"p90": 1.0,
"min": 1.0,
"max": 1.0
},
"speedup_of_means": 1.0,
"token_ratio": 1.0,
"sample_peak_alloc_gb": {
"n": 9,
"mean": 9.90966796875,
"std": 0.0,
"median": 9.90966796875,
"p10": 9.90966796875,
"p90": 9.90966796875,
"min": 9.90966796875,
"max": 9.90966796875
},
"sample_peak_reserved_gb": {
"n": 9,
"mean": 13.823567708333334,
"std": 0.001953125,
"median": 13.822265625,
"p10": 13.822265625,
"p90": 13.826171875,
"min": 13.822265625,
"max": 13.826171875
},
"sample_avg_alloc_gb": {
"n": 9,
"mean": 4.825455729166666,
"std": 0.005957831657292742,
"median": 4.82744140625,
"p10": 4.8201171875,
"p90": 4.82880859375,
"min": 4.81474609375,
"max": 4.83447265625
},
"sample_base_alloc_gb": {
"n": 9,
"mean": 4.50732421875,
"std": 0.0,
"median": 4.50732421875,
"p10": 4.50732421875,
"p90": 4.50732421875,
"min": 4.50732421875,
"max": 4.50732421875
},
"sample_peak_minus_base_per_obj_gb": {
"n": 9,
"mean": 0.67529296875,
"std": 0.0,
"median": 0.67529296875,
"p10": 0.67529296875,
"p90": 0.67529296875,
"min": 0.67529296875,