h3-blackwell-runtime/benchmarks/gb10-cute-p2-ring-capacity-attn-out-128.json
2026-08-25 20:30:22 +07:00

72 lines
1.5 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": [
128,
5376,
7168
],
"full_activation_rows": 37810,
"ring": {
"row_capacity": 128,
"producer": "vortex_native_quantize_nvfp4",
"qdata_bytes": 458752,
"sfa_bytes": 57344,
"logical_bytes": 516096
},
"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": 0.019654719829559325,
"global_scale_ms": 3.735771484375,
"cute_consumer_ms": 0.07179071903228759,
"modeled_ring_ms": 3.8272169232368465,
"actual_into_ring_cute_gemm_ms": 0.09406271934509278,
"comfy_consumer_ms": 0.05615776062011719,
"actual_into_ring_comfy_gemm_ms": 0.07875840187072754,
"reference_vortex_scale_comfy_pack_gemm_ms": 0.08742112159729004,
"modeled_ring_vs_reference": 43.77908740255154,
"actual_ring_vs_reference": 1.0759724609619812,
"comfy_ring_vs_reference": 0.9009081607707142
}
}