h3-blackwell-runtime/benchmarks/gb10-cute-p2-ring-capacity-attn-qkv-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": "attn_qkv_proj",
"mnk": [
2048,
21504,
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": 0.12289536476135254,
"global_scale_ms": 1.8337306213378906,
"cute_consumer_ms": 2.898988037109375,
"modeled_cute_chunk_ms": 3.0218834018707272,
"modeled_comfy_chunk_ms": 1.69343674659729,
"actual_into_ring_cute_gemm_ms": 1.5230323791503906,
"comfy_consumer_ms": 1.5705413818359375,
"actual_into_ring_comfy_gemm_ms": 1.7044099426269532,
"reference_vortex_scale_comfy_pack_gemm_ms": 1.7054486083984375,
"modeled_canonical_reference_ms": 34.2372541809082,
"modeled_canonical_cute_ring_ms": 30.77134582519531,
"modeled_canonical_comfy_ring_ms": 34.217519531250005,
"actual_ring_vs_reference": 0.8930391520742736,
"comfy_ring_vs_reference": 0.9993909721076499,
"modeled_canonical_cute_ring_vs_reference": 0.89876792287725,
"modeled_canonical_comfy_ring_vs_reference": 0.999423591344273
}
}