Skip to content
KernelIndex
Search⌘K
Sign in

Max reduction over a dimension

4 eligible runs
reduction

KernelBench level1 problem 49: Max reduction over a dimension. The computation is the reference PyTorch module's forward pass; the output shape follows the module (mirrored as each implementation's source).

Source baseline · unbeaten
3.14ms±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.

No package · vendor the source →View source →Run detail →

Current records

HardwareBest knownImplementationRuns
NVIDIA H100 · env 13.14 msPyTorch eager2NVIDIA H100 · env 24.75 msPyTorch eager2

Not measured on B200 for this workload. Challenges →

Source-native comparison · GPU NVIDIA H100 · Workload dim1 = 4096 · dim2 = 4095 · batch_size = 128 · fp32 · Protocol KernelBench timing scripts · mean · 2 results · last observed 2026-03-05Record history →
Estimated floor 2.56 ms · record 1.22× above itestimate, not evidence ›
DRAM 2.56 ms · bandwidth-bound on H100 SXM
every declared tensor crosses HBM exactly once (3,350 GB/s, H100 SXM datasheet)
no arithmetic formula for this family: bandwidth floor only
headroom-v1: a lower bound from declared tensors and datasheet peaks. A kernel can sit well above it for good reasons.
#
Implementation
Latency
vs #1
Trust
Observed
1
PyTorch eagerbaselinePyTorch
3.14ms±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 →
2
torch.compile (inductor)PyTorch
3.53ms±0.03
1.12×
Reported · MIT · source
2026-03-05stale

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

source mirroredMITno install recipeView source →Run detail →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
PyTorch eagerPyTorch
python · torch_eager
3.14ms
1.00×
Reported
MIT · source
torch.compile (inductor)PyTorch
python · torch_compile_inductor
3.53ms
1.12×
Reported
MIT · source

Semantics

Inputs and outputs
xfp32 [batch_size, dim1, dim2]
outfloat [out]
Axes and behavior
outvariable
dim1variable
dim2variable
batch_sizevariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
sha256258dbab896d0…
Sources: KernelBench baseline timings (2026-03-05) · MITlast observed 2026-03-05How records are decidedJSON