h3-blackwell-runtime/benchmarks/gb10-sage2-p0-pad-v-smem-latency.json

{
  "metadata": {
    "width": 1344,
    "height": 768,
    "frames": 124,
    "steps": 12,
    "sampler_step": 1,
    "seed": 440420,
    "text_tokens": 100,
    "block_index": 24,
    "attention": "sage2",
    "hidden_shape": [
      37810,
      5376
    ],
    "h_msa_shape": [
      37810,
      5376
    ],
    "segments": [
      [
        0,
        100,
        1
      ],
      [
        100,
        514,
        2
      ],
      [
        514,
        37810,
        0
      ]
    ]
  },
  "source_commit": "d1a57a546c3d395b1ffcbeecc66d81db76f3b4b5",
  "reference_output_sha256": "4c666c20f5f8f651158a2ced33ccff08f3bada07665c595b99008d171db30574",
  "q_shape": [
    1,
    37810,
    56,
    128
  ],
  "scheduler": {
    "cta_q": 128,
    "cta_k": 64,
    "warp_q": 32,
    "warp_k": 64,
    "threads_per_cta": 128,
    "dynamic_shared_memory_bytes": 32768,
    "grid_ctas": 16576
  },
  "rounds": 5,
  "warmup": 3,
  "iterations_per_round": 10,
  "variants": [
    {
      "name": "baseline",
      "path": "/variants/baseline/lib/sageattention/_qattn_sm89.cpython-312-aarch64-linux-gnu.so",
      "output_sha256": "4c666c20f5f8f651158a2ced33ccff08f3bada07665c595b99008d171db30574",
      "byte_exact": true,
      "timing": {
        "count": 50,
        "mean_s": 0.2461453451538086,
        "p50_s": 0.24636819458007814,
        "p90_s": 0.24946011962890624,
        "p95_s": 0.2502609535217285,
        "p99_s": 0.25086575302124026,
        "min_s": 0.24162757873535157,
        "max_s": 0.250990234375
      },
      "rounds": [
        {
          "count": 10,
          "mean_s": 0.2425358856201172,
          "p50_s": 0.24232412719726562,
          "p90_s": 0.2432921615600586,
          "p95_s": 0.2435843360900879,
          "p99_s": 0.24381807571411135,
          "min_s": 0.24162757873535157,
          "max_s": 0.2438765106201172
        },
        {
          "count": 10,
          "mean_s": 0.24422555541992189,
          "p50_s": 0.2443652801513672,
          "p90_s": 0.24503082122802733,
          "p95_s": 0.2450529930114746,
          "p99_s": 0.2450707304382324,
          "min_s": 0.24320451354980469,
          "max_s": 0.24507516479492186
        },
        {
          "count": 10,
          "mean_s": 0.24644686584472658,
          "p50_s": 0.24636819458007814,
          "p90_s": 0.24773657531738283,
          "p95_s": 0.24776796875,
          "p99_s": 0.24779308349609377,
          "min_s": 0.24484390258789063,
          "max_s": 0.2477993621826172
        },
        {
          "count": 10,
          "mean_s": 0.24803108673095703,
          "p50_s": 0.2479405059814453,
          "p90_s": 0.24909479675292967,
          "p95_s": 0.24924445800781247,
          "p99_s": 0.24936418701171875,
          "min_s": 0.2466182403564453,
          "max_s": 0.2493941192626953
        },
        {
          "count": 10,
          "mean_s": 0.24948733215332033,
          "p50_s": 0.24952471923828123,
          "p90_s": 0.2507615951538086,
          "p95_s": 0.2508759147644043,
          "p99_s": 0.25096737045288087,
          "min_s": 0.24757859802246093,
          "max_s": 0.250990234375
        }
      ],
      "speedup_vs_baseline_percent": 0.0
    },
    {
      "name": "pad-v-smem",
      "path": "/variants/pad-v-smem/lib/sageattention/_qattn_sm89.cpython-312-aarch64-linux-gnu.so",
      "output_sha256": "4c666c20f5f8f651158a2ced33ccff08f3bada07665c595b99008d171db30574",
      "byte_exact": true,
      "timing": {
        "count": 50,
        "mean_s": 0.24588236846923828,
        "p50_s": 0.2464697723388672,
        "p90_s": 0.24931472930908202,
        "p95_s": 0.24956529541015626,
        "p99_s": 0.25046807495117185,
        "min_s": 0.24145852661132813,
        "max_s": 0.25106903076171877
      },
      "rounds": [
        {
          "count": 10,
          "mean_s": 0.24199966735839845,
          "p50_s": 0.24201995086669922,
          "p90_s": 0.24234957427978515,
          "p95_s": 0.24245646438598634,
          "p99_s": 0.24254197647094727,
          "min_s": 0.24145852661132813,
          "max_s": 0.2425633544921875
        },
        {
          "count": 10,
          "mean_s": 0.24424241943359376,
          "p50_s": 0.24423397827148438,
          "p90_s": 0.2449089614868164,
          "p95_s": 0.24502979049682616,
          "p99_s": 0.245126453704834,
          "min_s": 0.2430936279296875,
          "max_s": 0.24515061950683595
        },
        {
          "count": 10,
          "mean_s": 0.24634171447753905,
          "p50_s": 0.2464697723388672,
          "p90_s": 0.2470756805419922,
          "p95_s": 0.2471369155883789,
          "p99_s": 0.24718590362548828,
          "min_s": 0.24513539123535155,
          "max_s": 0.24719815063476563
        },
        {
          "count": 10,
          "mean_s": 0.24810086364746092,
          "p50_s": 0.24811367034912107,
          "p90_s": 0.2494482360839844,
          "p95_s": 0.24964541320800782,
          "p99_s": 0.24980315490722654,
          "min_s": 0.2465597686767578,
          "max_s": 0.24984259033203124
        },
        {
          "count": 10,
          "mean_s": 0.2487271774291992,
          "p50_s": 0.24817182159423828,
          "p90_s": 0.2497320083618164,
          "p95_s": 0.25040051956176756,
          "p99_s": 0.25093532852172856,
          "min_s": 0.24757408142089843,
          "max_s": 0.25106903076171877
        }
      ],
      "speedup_vs_baseline_percent": -0.041213069588663664
    }
  ]
}