Inspector 2.3.0.13

Purpose & User Intent: This tool provides local, read-only inspection of NVIDIA GPU hardware and driver state. It targets developers, QA engineers, IT admins, and power users who need rapid visibility into GPU health for troubleshooting, validation, and benchmarking. It is used when diagnosing driver crashes, overheating, performance regressions, or sensor anomalies during hardware validation and field support.
Core Logic & Features: The tool queries GPU sensors via native APIs to collect model, driver version, temperature, clock speeds, memory usage, fan speed, and power draw. It supports multiple GPUs, timestamped records, and export of structured data. Required features include read-only sensor access and cross-GPU enumeration; optional features include configurable polling intervals and CSV/JSON exports.
Inputs & Outputs & Algorithms: Inputs include optional scope (gpu0, all), poll_interval in seconds, and export_format (csv/json). Outputs are a snapshot object containing metrics, timestamp, and optional export_path. Algorithms involve sensor data normalization, unit conversion (Celsius ↔ Fahrenheit if needed), and simple health scoring. Outputs are machine-readable, suitable for pipelines and telemetry.
Error & Edge Cases; Industry/Region & Localization; Assumptions & Exclusions: Handles missing sensors by reporting unavailable fields as null, and returns a partial snapshot. If multiple GPUs exist, it enumerates all devices. When drivers are not loaded or permissions are insufficient, the tool raises a controlled error state without blocking other operations. Timeouts and invalid formats default to safe fallbacks. Uses standard GPU sensor units (°C, MHz, W) and remains English-localized; excludes overclocking guidance and remote monitoring features.