h3-blackwell-runtime/benchmarks/gb10-cute-p2-ring-capacity-attn-out-4096.json
2026-08-25 20:30:22 +07:00

78 lines
1.9 KiB
JSON

{
"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": "attn_out_proj",
"mnk": [
4096,
5376,
7168
],
"full_activation_rows": 37810,
"modeled_chunk_count": 10,
"ring": {
"row_capacity": 4096,
"producer": "vortex_native_quantize_nvfp4",
"qdata_bytes": 14680064,
"sfa_bytes": 1835008,
"logical_bytes": 16515072
},
"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": 50,
"producer_ms": 0.3313638305664062,
"global_scale_ms": 2.197855987548828,
"cute_consumer_ms": 1.3521932983398437,
"modeled_cute_chunk_ms": 1.6835571289062499,
"modeled_comfy_chunk_ms": 1.7651193237304688,
"actual_into_ring_cute_gemm_ms": 1.8329510498046875,
"comfy_consumer_ms": 1.4337554931640626,
"actual_into_ring_comfy_gemm_ms": 1.6257913208007813,
"reference_vortex_scale_comfy_pack_gemm_ms": 1.596439666748047,
"modeled_canonical_reference_ms": 18.1622526550293,
"modeled_canonical_cute_ring_ms": 20.527366485595703,
"modeled_canonical_comfy_ring_ms": 18.45576919555664,
"actual_ring_vs_reference": 1.148149277409534,
"comfy_ring_vs_reference": 1.0183856957854998,
"modeled_canonical_cute_ring_vs_reference": 1.130221392438976,
"modeled_canonical_comfy_ring_vs_reference": 1.0161608004305602
}
}