{
    "mode": "1 GPU, matmul overlapped with a 64 MiB copy engine transfer",
    "backend": "cuda streams",
    "worlds": [
        1
    ],
    "checks": 9,
    "max_abs_vs_reference": 1.9073486328125e-05,
    "max_abs_vs_fp64": 7.273328086920827e-06,
    "max_abs_copy": 0.0,
    "fp64_rows_per_seed": 64,
    "reference_method": "unchunked a @ b vs row chunked a @ b under a side stream copy, sampled rows vs a float64 CPU product",
    "shape": {
        "m": 4096,
        "k": 4096,
        "n": 4096
    },
    "timing_shape": {
        "m": 4096,
        "k": 4096,
        "n": 4096
    },
    "timing_ms": {
        "compute": 35.353,
        "comm": 5.141,
        "reference_step": 37.872,
        "overlapped_step": 35.863,
        "exposed_comm_reference": 2.518,
        "exposed_comm_overlapped": 0.51
    },
    "timing_world": 1,
    "sample": "fused-allreduce-overlap",
    "device": "cuda",
    "torch": "2.11.0+cu130",
    "host": "Linux x86_64",
    "atol": 0.001,
    "fp64_bound": 0.001,
    "seeds": [
        0,
        1,
        2
    ],
    "chip": "Tesla T4 x1",
    "arch": "sm_75",
    "cuda": "13.0",
    "passed": true
}