Skip to content
KernelIndex
Search⌘K
Sign in

conv depthwise separable 2D

4 eligible runs
conv

KernelBench level1 problem 86: conv depthwise separable 2D. 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
3.91ms±0.36 · 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 13.91 msPyTorch eager2NVIDIA H100 · env 25.18 msPyTorch eager2

Not measured on B200 for this workload. Challenges →

Source-native comparison · GPU NVIDIA H100 · Workload width = 512 · height = 512 · batch_size = 16 · in_channels = 64 · fp32 · Protocol KernelBench timing scripts · mean · 2 results · last observed 2026-03-05Record history →
Estimated floor 320.5 µs · record 12.2× above itestimate, not evidence ›
DRAM 320.5 µs · 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
3.91ms±0.36
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
4.18ms±0.02
1.07×
Reported · MIT · source
2026-03-05stale

1.07× 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
3.91ms
1.00×
Reported
MIT · source
torch.compile (inductor)PyTorch
python · torch_compile_inductor
4.18ms
1.07×
Reported
MIT · source

Semantics

Inputs and outputs
xfp32 [batch_size, in_channels, height, width]
outfloat [out]
Axes and behavior
outvariable
widthvariable
heightvariable
batch_sizevariable
in_channelsvariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
sha25672521e2eb6a8…
Sources: KernelBench baseline timings (2026-03-05) · MITlast observed 2026-03-05How records are decidedJSON