h3-blackwell-runtime/benchmarks/gb10-cute-p2-ring-capacity-mlp-fc1-2048.json

79 lines
1.9 KiB
JSON
Raw Normal View History

{
"device": "NVIDIA GB10",
"cutlass_dsl": "4.6.2",
"metadata": {
"width": 1344,
"height": 768,
"frames": 124,
"steps": 12,
"sampler_step": 1,
"seed": 440420,
"text_tokens": 100,
"hidden_shape": [
37810,
5376
],
"segments": [
[
0,
100,
1
],
[
100,
514,
2
],
[
514,
37810,
0
]
]
},
"linear": "mlp_fc1",
"mnk": [
2048,
28672,
5376
],
"full_activation_rows": 37810,
"modeled_chunk_count": 19,
"ring": {
"row_capacity": 2048,
"producer": "vortex_native_quantize_nvfp4",
"qdata_bytes": 5505024,
"sfa_bytes": 688128,
"logical_bytes": 6193152
},
"parity": {
"tensor_scale_equal": true,
"fp4_difference_count": 0,
"block_scale_difference_count": 0,
"output_equal": true,
"max_abs": 0.0,
"mean_abs": 0.0,
"comfy_output_equal": true
},
"timing": {
"warmup": 10,
"iterations": 100,
"producer_ms": 0.13199872016906739,
"global_scale_ms": 2.422532806396484,
"cute_consumer_ms": 6.302136840820313,
"modeled_cute_chunk_ms": 6.434135560989381,
"modeled_comfy_chunk_ms": 2.247994508743286,
"actual_into_ring_cute_gemm_ms": 2.0069894409179687,
"comfy_consumer_ms": 2.1159957885742187,
"actual_into_ring_comfy_gemm_ms": 2.2663865661621094,
"reference_vortex_scale_comfy_pack_gemm_ms": 2.26166748046875,
"modeled_canonical_reference_ms": 45.394214935302735,
"modeled_canonical_cute_ring_ms": 40.55533218383789,
"modeled_canonical_comfy_ring_ms": 45.48387756347656,
"actual_ring_vs_reference": 0.8873936855218004,
"comfy_ring_vs_reference": 1.0020865515086157,
"modeled_canonical_cute_ring_vs_reference": 0.8934030964438668,
"modeled_canonical_comfy_ring_vs_reference": 1.0019751994456036
}
}