workloads.md

layout	default
title	Workload Explorer
permalink	/workloads/
description	Explore AI workloads and evidence-scoped hardware/software execution paths.

MODEL × WORKLOAD × SYSTEM

Workload Explorer

Start with the workload, not the chip. Explore candidate execution paths and see exactly what the available evidence establishes — and what remains unverified.

Dataset Loading… Last reviewed — Raw workload data ↗ JSON Schema ↗

Workload definitions—

Candidate execution paths—

Integration documented only—

Vendor-reported—

DEFINE YOUR WORKLOAD

Find candidate execution paths

Reset filters

Search Workload All workloads Accelerator All platforms Evidence status All statuses

CANDIDATE PATHS

— paths

Loading workload data…

Loading workload data…

No matching paths

Try a broader query or reset the filters.

WORKLOAD TAXONOMY

What should be measured?

Each definition identifies the workload dimensions that must be pinned before results can be compared.

EVIDENCE POLICY

Integration is not validation.

Integration only means the software-to-platform path is documented. Vendor reported means a vendor has described an adaptation. Neither establishes a reproducible workload test. Higher maturity requires an explicit model, device configuration, software versions, correctness criteria and, for benchmark claims, a measurement methodology and raw result evidence. Empty metric arrays are intentional — no performance numbers are fabricated.

<script src="{{ '/assets/js/workload-explorer.js' | relative_url }}" defer></script>