PyTorch eager
ScaledDotProductAttention · num_heads = 32 · batch_size = 32 · sequence_length = 512 · embedding_dimension = 1024 · fp32 · run 01a03f0e-f837…
●passedReported evidence
Not re-observed recently.
Primary measurement
13.3ms±0.12 · mean of 100
Rank 2 in its comparison group · source-native comparison · observed 2026-03-05
max ms 14.3 · min ms 12.1 · std ms 0.602 · mean ms 13.3
Identity
implementationPyTorch eager
projectPyTorch
revisionunknown
workloadnum_heads = 32 · batch_size = 32 · sequence_length = 512 · embedding_dimension = 1024 · fp32
comparison keysha256:7edf468fa7ecd7b2…
sourceKernelBench baseline timings
external idH100_PCIe_LambdaLabs/torch/level1/97_ScaledDotProductAttention.py
sha256:9f41183f7bfec606a2d5c57…
Correctness
Marked passed by the source; the correctness policy was not published.
Workload
num_heads32
batch_size32
sequence_length512
embedding_dimension1024
kfp32 [32, 32, 512, 1024]
qfp32 [32, 32, 512, 1024]
vfp32 [32, 32, 512, 1024]
definition comparatornot_asserted
Measurements
latency · max14.3 ms · n=100
latency · mean13.3 ms · n=100
latency · min12.1 ms · n=100
latency · std602.0 µs
Protocol
harnessKernelBench timing scripts
timercuda_events
primaryStatisticmean
comparabilityFamilykernelbench_baseline_timing
Environment
gpuNVIDIA H100 (sm_90)
Artifacts
No artifacts published with this run.
Replications and notes
No attestations yet.
Community attestations. They never change the evidence level; only a KernelIndex-controlled rerun does.
Add a reproduction or note
Canonical manifest
Show manifestHide manifest
{
"run": {
"kind": "BenchmarkRun",
"spec": {
"status": "passed",
"timing": {
"samples": 100,
"latencyNs": {
"mean": 13300000,
"maximum": 14300000,
"minimum": 12100000,
"confidence95": [
13182008,
13417992
]
},
"primaryStatistic": "mean"
},
"observedAt": "2026-03-05T08:38:15.000Z",
"measurements": [
{
"unit": "ns",
"value": 602000,
"metric": "latency",
"statistic": "std"
}
],
"sourceNative": {
"source": "kernelbench",
"metrics": {
"max_ms": 14.3,
"min_ms": 12.1,
"std_ms": 0.602,
"mean_ms": 13.3
},
"benchmark": "H100_PCIe_LambdaLabs/baseline_time_torch.json",
"externalId": "H100_PCIe_LambdaLabs/torch/level1/97_ScaledDotProductAttention.py"
},
"protocolDigest": "sha256:547725b033a9911eae0c5ac362344a85aed824894ed4ab476af945b9bb491ea6",
"workloadDigest": "sha256:2467b1ffd57daa52d0649a4267cc28cd096cf62c1b9cdf4e2171bcb0dd0a2eb3",
"environmentDigest": "sha256:89f9b28b06758f22a1df331fea58bfde2c9cc0f3627d1829509743ea538f0011",
"implementationDigest": "sha256:ac42fea4be49f0f5330f1ddf18aac4530509bdd9f7cb3f642943d0e60c3aa6bf"
},
"metadata": {
"name": "kernelbench-h100-pcie-lambdalabs-torch-l1-97-scaleddotproductattention",
"title": "ScaledDotProductAttention · PyTorch eager · Lambda Labs"
},
"apiVersion": "kernelindex.dev/v1alpha1"
},
"protocol": {
"kind": "BenchmarkProtocol",
"spec": {
"harness": {
"name": "KernelBench timing scripts",
"repository": "https://github.com/ScalingIntelligence/KernelBench"
},
"measurement": {
"timer": "cuda_events",
"primaryStatistic": "mean"
},
"comparability": {
"notes": "Mean of 100 timed forward passes (CUDA events, warm-up excluded) of the reference module on fixed inputs; the 95% interval is the mean's, from the reported standard deviation. The upstream JSON records no torch or CUDA version. Comparable only within one problem and one timing host.",
"family": "kernelbench_baseline_timing"
}
},
"metadata": {
"name": "kernelbench-timing-v1",
"title": "KernelBench baseline timing"
},
"apiVersion": "kernelindex.dev/v1alpha1"
},
"environment": {
"kind": "ExecutionEnvironment",
"spec": {
"hardware": {
"vendor": "nvidia",
"product": "NVIDIA H100",
"formFactor": "PCIe",
"architecture": "sm_90"
},
"software": {}
},
"metadata": {
"name": "kernelbench-h100-pcie-lambdalabs",
"title": "KernelBench host · H100 PCIe (Lambda Labs)"
},
"apiVersion": "kernelindex.dev/v1alpha1"
}
}Cite this record (permalink, digest, access date)
Report an issue with this run
Published 2026-08-26 · KernelBench baseline timings · MITAll results for ScaledDotProductAttention →JSON