h3-blackwell-runtime/benchmarks/gb10-cute-p2-ring-capacity-attn-out-1024.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": "attn_out_proj",
"mnk": [
1024,
5376,
7168
],
"full_activation_rows": 37810,
"modeled_chunk_count": 37,
"ring": {
"row_capacity": 1024,
"producer": "vortex_native_quantize_nvfp4",
"qdata_bytes": 3670016,
"sfa_bytes": 458752,
"logical_bytes": 4128768
},
"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": 0.08987903594970703,
"global_scale_ms": 4.179701232910157,
"cute_consumer_ms": 0.44698047637939453,
"modeled_cute_chunk_ms": 0.5368595123291016,
"modeled_comfy_chunk_ms": 0.5336019134521484,
"actual_into_ring_cute_gemm_ms": 0.5717164611816407,
"comfy_consumer_ms": 0.4437228775024414,
"actual_into_ring_comfy_gemm_ms": 0.6249113464355469,
"reference_vortex_scale_comfy_pack_gemm_ms": 0.5978803253173828,
"modeled_canonical_reference_ms": 26.301273269653322,
"modeled_canonical_cute_ring_ms": 25.33321029663086,
"modeled_canonical_comfy_ring_ms": 27.301421051025393,
"actual_ring_vs_reference": 0.9562389611635855,
"comfy_ring_vs_reference": 1.0452114243829898,
"modeled_canonical_cute_ring_vs_reference": 0.9631933038717398,
"modeled_canonical_comfy_ring_vs_reference": 1.0380265917591927
}
}