Skip to content
KernelIndex
Search⌘K

Geometric

GQA paged prefill causal h32 kv8 d128 ps1 · suite of 38 cases · mean latency · run 01a0032e-d8e9…
●passedReported evidence
Primary measurement
5.18µsmean

Rank 3 in its comparison group · source-native comparison · observed 2026-07-28

Compare with #1 →

SOL 0.9889 · 127.96× reference · 38/38 cases faster

Identity
implementationGeometric
projectGeometric
revisionsubmission-30229
workloadsuite of 38 cases · mean latency
comparison keysha256:a204cd732f8b97c7…
sourceNVIDIA SOL-ExecBench
external idsubmission/30229
sha256:6faa87a0fc7db9d9c96566b…

Correctness

Marked passed by the source; the correctness policy was not published.

Workload

definition comparatorsol_execbench_eval
descriptionCorrectness must pass on every case in the SOL evaluation stack.

Measurements

latency · mean5.18 µs

Protocol

harnessSOL-ExecBench evaluation stack v1.1
timerharness_reported
primaryStatisticmean
correctnessReferencedefinition reference implementation
comparabilityFamilysol_execbench_leaderboard

Environment

gpuNVIDIA B200 (sm_100)

Artifacts

No artifacts published with this run.

Replications and notes

No attestations yet.

Community attestations. They never change the evidence level; only a KernelIndex-controlled rerun does.

Add a reproduction or note
I

Published under your account name. Community attestations never change the evidence level; a maintainer can hide abuse.

Canonical manifest

Show manifest
{
  "run": {
    "kind": "BenchmarkRun",
    "spec": {
      "status": "passed",
      "timing": {
        "latencyNs": {
          "mean": 5180
        },
        "primaryStatistic": "mean"
      },
      "observedAt": "2026-07-28T18:03:09.953Z",
      "sourceNative": {
        "source": "sol-execbench",
        "metrics": {
          "sol_score": 0.988933,
          "latency_ms": 0.00518,
          "avg_speedup": 127.9611,
          "fast_1_count": 38,
          "fast_1_total": 38
        },
        "benchmark": "015_gqa_paged_prefill_causal_h32_kv8_d128_ps1",
        "externalId": "30229"
      },
      "protocolDigest": "sha256:6383f62490dad270ad2146a94f67b48fba5cfcbc0b9b180f02665c69cc4433a0",
      "workloadDigest": "sha256:5139faeaf6adf6ccb44b2d9567a45e470eec1ed17e187a3e094c743b9b972eed",
      "environmentDigest": "sha256:f304afb2fedab4bbeadf8ac36a990ff517ab6e824dfe5a3ee0fe46259ff5e39f",
      "implementationDigest": "sha256:920c84407b8723f6f68d48651a0a68a5e0f71cd7fc6f97b8e702a905d803237a"
    },
    "metadata": {
      "name": "sol-submission-30229",
      "title": "Geometric · 015_gqa_paged_prefill_causal_h32_kv8_d128_ps1 · B200",
      "labels": {
        "worker": "external-prod-b200-modal-1"
      }
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  },
  "protocol": {
    "kind": "BenchmarkProtocol",
    "spec": {
      "harness": {
        "name": "SOL-ExecBench evaluation stack",
        "version": "v1.1"
      },
      "correctness": {
        "reference": "definition reference implementation",
        "comparator": "sol_execbench_eval"
      },
      "measurement": {
        "timer": "harness_reported",
        "primaryStatistic": "mean"
      },
      "comparability": {
        "notes": "Suite-mean latency and SOL-Score as published by the SOL-ExecBench leaderboard. Comparable only within one definition, GPU type, and evaluation stack version.",
        "family": "sol_execbench_leaderboard"
      }
    },
    "metadata": {
      "name": "sol-execbench-leaderboard-v1-1",
      "title": "SOL-ExecBench evaluation stack v1.1"
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  },
  "environment": {
    "kind": "ExecutionEnvironment",
    "spec": {
      "hardware": {
        "vendor": "nvidia",
        "product": "NVIDIA B200",
        "architecture": "sm_100"
      },
      "software": {
        "libraries": {
          "evaluation_stack": "v1.1"
        }
      }
    },
    "metadata": {
      "name": "sol-execbench-b200-v1-1",
      "title": "SOL-ExecBench NVIDIA B200, stack v1.1"
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  }
}
Cite this record (permalink, digest, access date)
Report an issue with this run

A maintainer reviews every report. If accepted, the record is retracted or superseded; its history stays visible.

Published 2026-08-15 · NVIDIA SOL-ExecBenchAll results for GQA paged prefill causal h32 kv8 d128 ps1 →JSON