Workload-classified GPU fleet telemetry

FARMGPUA5 · direct accessManage supplier listing

Workload-classified GPU fleet telemetry: a catalog of 50,000+ distinct GPU workloads over a 90-day window, each a ~100-field profile joining hardware telemetry (utilisation, memory, power draw, temperature, NVLink, ECC error counts) with container-level metrics and an LLM-assigned semantic workload label derived from image name, metric shape, and GitHub/Docker Hub repo lookup. Uniquely, it pairs measured utilisation with what is actually running and why — a demand-composition and efficiency asset rather than another price feed — plus a quantified waste layer (GPU-hours burned below 10 percent utilisation, worth $500K+ annually on the reference fleet) and an observed workload-mix distribution across development, image generation, inference, video generation, fine-tuning and training.

Open buyer room →

Sample

Brickroad customers

Sample this dataset before you buy it.

Your sourcing agent asks Farmgpu and files the sample in your catalog — private to you.

Start 7-day trialNothing is charged until the trial ends.

Coverage

Universe, instruments and categories.

Industry: IT Consulting & Other ServicesFinancial Exchanges & Data
Instruments: equitiesfixed_income
Categories: AlternativeSupply ChainGeospatial
Regions: US
Sample tickers: NVDAAMDCRWVNBISIRENAPLDSMCIVRTETNPWR

Dataset card

Type: not stated
Format: Multimodal — structured time-series telemetry catalogued in Postgres, exportable as CSV/JSON, with Prometheus/Thanos query access and a CLI
Volume: 50,000+ cataloged workloads × ~100 fields per profile ≈ 5M+ field-level observations over 90 days, across 5,000+ unique container images; underlying DCGM + cAdvisor + Node Exporter time series at Prometheus scrape resolution (typically 15-60s) per GPU per node, so the raw metric volume is orders of magnitude larger than the catalogued profile layer; plus 11 dashboard views and CSV/JSON export
Users: No consumer registration surface exists — an internal fleet-intelligence platform run by a GPU cloud over its own and its customers' clusters, so the relevant count is nodes and workloads under management rather than registrants.
History: not stated
Update frequency: not stated
Growth: Active — an incremental build pipeline that processes only new data since last run (indicating continuous ingestion rather than a one-off export), a shipped optimisation reducing classification calls by 85%, and a mature tooling surface (11 dashboards, an 11-command CLI, CLI export) maintained by a 15-person company
Launched: est. 2025 (no launch date published). Bounded by a demonstrated 90-day cataloging window already accumulated and the presence of an optimised, matured pipeline (85% classification-call reduction implies iteration beyond a first version); the operating company records 15 employees and no disclosed funding.
Delivery: not stated
Entity mapping: not stated
Sample: not stated
Point-in-time: not stated
Licence: not stated

supplier index read from their site

Farmgpu has not added their own details yet. Not yet on file:

Dataset
Data dictionary
Coverage
Provenance
Rights & data handling