Search · KernelIndex

Fastest reported

7.10µsmean · 89% of SOL

QwenLicense unknown · unknown

Reported evidence · last observed 2026-07-23. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Faster under a different protocol, not comparable here: 6.64 µs View →

Ranked under the source's published protocolWhy comparable?

GPUNVIDIA B200

Workloadsuite of 14 cases · mean latency

ProtocolSOL-ExecBench evaluation stack v1.1 · mean

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

Exact 36 Compatible 112 Supported 1 Other protocols 17

Has source1 License known1Installable0

Claims support; no run on this exact workload yet.

Implementation

Latency

Trust

Observed

—

gemini-2.5-pro / cuda5808cdgemini-2.5-pro

—

— · Apache-2.0 · source

—

Declared support only; no measurement for this workload.

source mirroredApache-2.0no install recipeView source →

RMSNormOperation in the rmsnorm familyFused add RMSNorm h2048Operation in the rmsnorm familyFused add RMSNorm h3072Operation in the rmsnorm familyFused add RMSNorm h5120Operation in the rmsnorm familyFused add RMSNorm h5376Operation in the rmsnorm familyFused add RMSNorm h7168Operation in the rmsnorm family

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

This search as an API call

curlkicomparison key sha256:8ee21acd086e6c19…JSON

FlashInfer-Bench · last observed 2025-10-21 · NVIDIA SOL-ExecBench · last observed 2026-08-21