Skip to content
KernelIndex
Search⌘K
Sign in

GEMM n5120 k17408

129 eligible runs
gemm

General matrix multiply (GEMM) C = A @ B.T. Captured from Qwen3 14B down_proj (intermediate=17408 → hidden=5120).

Source baseline · unbeaten
41.2µsmean
flashinfer / wrapperdd130aFlashInfer-Bench baselines · Apache-2.0 · python

Reported evidence · last observed 2026-03-24. The source's designated baseline implementation. Reported by source; not independently reproduced.

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

Current records

Workloadm = 4 · fp1643 cases
m
dtype
1fp162fp164fp167fp168fp1615fp1616fp1624fp1632fp1635fp1640fp1648fp1656fp1664fp1670fp1672fp1680fp1688fp1696fp16104fp16112fp16120fp16128fp16136fp16144fp16152fp16160fp16168fp16176fp16184fp16192fp16200fp16208fp16216fp16224fp16232fp16240fp16248fp16256fp16972fp162053fp162379fp168192fp16
scrolls · 43 cases total

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload m = 4 · fp16 · CUDA 12.8 · Framework pytorch 2.9.1+cu128 · Protocol flashinfer-bench · mean · 3 results · last observed 2026-03-24Record history →
Estimated floor 22.3 µs · record 1.85× above itestimate, not evidence ›
DRAM 22.3 µs · compute 317 ns · bandwidth-bound on B200
every declared tensor crosses HBM exactly once (8,000 GB/s, B200 datasheet)
2·M·N·K with M=4, N=5120, K=17408 at the dense bf16 peak (2,250 TFLOP/s)
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
flashinfer / wrapperdd130abaselineFlashInfer-Bench baselines
41.2µs
1.00×
Reported · Apache-2.0 · source
2026-03-24stale

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

source mirroredApache-2.0no install recipeView source →Run detail →
2
flashinfer / wrapperdd130abaselineFlashInfer-Bench baselines
42.1µs
1.02×
Reported · Apache-2.0 · source
2026-03-23stale

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

source mirroredApache-2.0no install recipeView source →Run detail →
3
flashinfer / wrapperdd130abaselineFlashInfer-Bench baselines
42.2µs
1.02×
Reported · Apache-2.0 · source
2026-03-24stale

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

source mirroredApache-2.0no install recipeView source →Run detail →

Scaling by m

100.0 µs1.00 ms14248825620538km →42.0 µs42.2 µs41.2 µs41.8 µs41.7 µs48.2 µs42.4 µs47.7 µs42.1 µs50.7 µs50.7 µs52.2 µs50.0 µs45.9 µs50.0 µs49.3 µs44.5 µs49.8 µs44.9 µs44.4 µs44.6 µs44.8 µs45.0 µs58.2 µs58.1 µs54.9 µs61.2 µs58.3 µs61.0 µs61.5 µs61.3 µs55.0 µs54.7 µs53.8 µs53.7 µs53.6 µs53.8 µs53.5 µs53.8 µs120.9 µs244.0 µs295.7 µs961.8 µsflashinfer / wrapper dd130a
flashinfer / wrapper dd130abest per workload · NVIDIA B200 · CUDA 12.8 · pytorch 2.9.1+cu128 · flashinfer-bench held constant · log scale

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
flashinfer / wrapperdd130aFlashInfer-Bench baselines
python
41.2µs
1.00×
Reported
Apache-2.0 · source

Semantics

Inputs and outputs
afp16 [m, k]
bfp16 [n, k]
cfp16 [m, n]
Axes and behavior
kconstant = 17408
mvariable
nconstant = 5120
determinismunspecified
constraintsNo mutation or aliasing
Identity
aliasgemm_n5120_k17408modelqwen3-14bsha2563c9ed3c3712a…
Sources: FlashInfer-Bench (2026-03-24) · Apache-2.0last observed 2026-03-24How records are decidedJSON