{
  "title": "PyC latest kernel suite",
  "updated_utc": "2026-04-21T20:22:52Z",
  "campaign": "hopper_box_20260421",
  "hardware_class": "Hopper SM90",
  "rows": [
    {
      "kernel": "ada_tensor_core_fp16",
      "timestamp_utc": "2026-04-21T19:17:25.053119+00:00",
      "arch": null,
      "shape": "1024x1024x1024",
      "precision": "fp16",
      "best_ms": 0.134,
      "tflops": null,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-tensorcore-bringup-20260421T191817Z.json"
    },
    {
      "kernel": "hopper_cublaslt_bf16",
      "timestamp_utc": "2026-04-21T20:22:52.293855+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 0.161,
      "tflops": 853.022,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-sm90-gap-close-20260421T202304Z.json"
    },
    {
      "kernel": "hopper_cublaslt_bf16_check",
      "timestamp_utc": "2026-04-21T19:45:45.425228+00:00",
      "arch": "sm90",
      "shape": "512x512x512",
      "precision": "bf16",
      "best_ms": 0.009,
      "tflops": 30.615,
      "source_artifact": "./results/kernel-suite/artifacts/hopper_cublaslt_bf16_check-20260421T194551Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16",
      "timestamp_utc": "2026-04-21T19:18:04.354083+00:00",
      "arch": "sm90",
      "shape": "1024x1024x1024",
      "precision": "bf16",
      "best_ms": 0.102,
      "tflops": 21.024,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-tensorcore-bringup-20260421T191817Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_async",
      "timestamp_utc": "2026-04-21T20:11:24.498547+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 1.059,
      "tflops": 129.824,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-sm90-80-120-async-candidates-20260421T201129Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_async_square",
      "timestamp_utc": "2026-04-21T20:23:00.201067+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 0.967,
      "tflops": 142.119,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-sm90-gap-close-20260421T202304Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_async_square_k64",
      "timestamp_utc": "2026-04-21T20:23:02.603962+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 0.926,
      "tflops": 148.394,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-sm90-gap-close-20260421T202304Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_async_wide",
      "timestamp_utc": "2026-04-21T20:22:55.397333+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 1.004,
      "tflops": 136.839,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-sm90-gap-close-20260421T202304Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_async_wide_k64",
      "timestamp_utc": "2026-04-21T20:22:57.797637+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 1.016,
      "tflops": 135.326,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-sm90-gap-close-20260421T202304Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_k32",
      "timestamp_utc": "2026-04-21T19:38:11.517286+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 6.506,
      "tflops": 21.125,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-bf16-nextloop-wmma-20260421T193831Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_perf",
      "timestamp_utc": "2026-04-21T20:11:18.323287+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 5.118,
      "tflops": 26.855,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-sm90-80-120-async-candidates-20260421T201129Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_warp2n",
      "timestamp_utc": "2026-04-21T19:38:01.568355+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 5.571,
      "tflops": 24.672,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-bf16-nextloop-wmma-20260421T193831Z.json"
    },
    {
      "kernel": "hopper_tensor_core_bf16_warp2n_k32",
      "timestamp_utc": "2026-04-21T19:38:21.657799+00:00",
      "arch": "sm90",
      "shape": "4096x4096x4096",
      "precision": "bf16",
      "best_ms": 6.112,
      "tflops": 22.486,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-bf16-nextloop-wmma-20260421T193831Z.json"
    },
    {
      "kernel": "hopper_tensor_core_fp16",
      "timestamp_utc": "2026-04-21T19:17:44.838086+00:00",
      "arch": "sm90",
      "shape": "1024x1024x1024",
      "precision": "fp16",
      "best_ms": 0.105,
      "tflops": 20.473,
      "source_artifact": "./results/kernel-suite/artifacts/hopper-tensorcore-bringup-20260421T191817Z.json"
    }
  ],
  "plots": [
    {
      "label": "Async square K64 kernel flamegraph",
      "path": "./results/kernel-suite/artifacts/hopper_tensor_core_bf16_async_square_k64_4096_flame.svg"
    },
    {
      "label": "Async square K64 stall flamegraph",
      "path": "./results/kernel-suite/artifacts/hopper_tensor_core_bf16_async_square_k64_4096_stall_flamegraph.svg"
    },
    {
      "label": "Async square K64 stall pyramid",
      "path": "./results/kernel-suite/artifacts/hopper_tensor_core_bf16_async_square_k64_4096_stall_pyramid.svg"
    },
    {
      "label": "cuBLASLt BF16 control flamegraph",
      "path": "./results/kernel-suite/artifacts/hopper_cublaslt_bf16_4096_flame.svg"
    }
  ],
  "notes": [
    "Rows are latest observed kernel-reported measurements per kernel name from the April 21 campaign.",
    "Kernel-reported best timing and TFLOP/s are not end-to-end inference throughput.",
    "Profiler artifacts are provided for investigation; compare only like-for-like shapes and precision."
  ]
}