Search · KernelIndex

Fastest reported

3.24ms±0.00 · mean of 100

PyTorch eagerPyTorch · MIT · python

Reported evidence · last observed 2026-03-05. The source's designated baseline implementation. Reported by source; not independently reproduced.

Ranked under the source's published protocolWhy comparable?

GPUNVIDIA H100

Workloaddim1 = 4096 · dim2 = 4095 · batch_size = 128 · fp32

ProtocolKernelBench timing scripts · mean

The query left these dimensions open; this is the most-measured of 2 cohorts. The hardware chips above switch it.

Implementation

Latency

vs #1

Trust

Observed

PyTorch eagerbaselinePyTorch

3.24ms±0.00

1.00×

Reported · MIT · source

2026-03-05stale

Measured exactly what you asked. The source's designated baseline implementation. Reported by source; not independently reproduced.

source mirroredMITno install recipeView source →Run detail →

3.37ms±0.06

1.04×

Reported · MIT · source

2026-03-05stale

1.04× slower than the baseline. Measured exactly what you asked. Reported by source; not independently reproduced.

Vector sum reductionOperation in the reduction familyMax Pooling 1DOperation in the reduction familyMax Pooling 2DOperation in the reduction familyMax Pooling 3DOperation in the reduction familySum reduction over a dimensionOperation in the reduction familyMean reduction over a dimensionOperation in the reduction family

Missing a kernel here? Evidence submissions open with the contribution beta.

This search as an API call

curlkicomparison key sha256:89733b04984de42f…JSON

KernelBench baseline timings · last observed 2026-03-05

Argmin over a dimension