GPU Scout: Real-Time Cloud GPU Monitoring with Spot Price Alerts
NEO built GPU Scout to monitor GPU availability and pricing across major cloud and ML-focused providers, alerting when spot prices drop or capacity opens up so teams can plan training runs strategically.
Problem Statement
We asked NEO to pull scattered GPU markets into one feed: prices, regions, and what is actually available, plus alerts when price or capacity crosses the lines you care about.
Solution Overview
- Multi-provider polling: AWS, GCP, Azure, Lambda Labs, and similar ML marketplaces.
- Normalization: One schema for GPU type, region, price per hour, and availability.
- Alerts: Slack, email, webhooks, and hooks if you want automation on top.
Workflow / Pipeline
| Step | Description |
|---|---|
| 1. Poll | Scheduled fetch per provider adapter |
| 2. Normalize | Map to unified GPU and region keys |
| 3. Store | Time-series DB for history and dashboards |
| 4. Alert | Evaluate rules; notify channels on triggers |
Repository & Artifacts
Generated Artifacts:
- Monitoring daemon and provider plugins
- Dashboards for price history and provider comparison