About this app
Local GPU activity, with context.
RTX AI Activity Monitor brings live GPU readings, a quiet tray indicator and local activity history into one focused Windows utility. Review available GPU load, dedicated memory, temperature and power readings. Choose an hour, day, week or custom history interval, then export the selected GPU's recorded samples and completed app-reported sessions as CSV or JSON. You can also save a PNG of the current app window.
The activity labels explain different kinds of evidence:
• Confirmed (app-reported): a cooperating application reports an active session through the local integration. The monitor verifies the connecting process ID, but does not independently prove its computations or GPU use.
• Probable (estimated): a configured process name matches correlated GPU activity that meets the chosen threshold. This is a heuristic, not certain AI detection.
• Unknown: readings are unavailable or observed activity cannot be classified reliably.
• Idle: measured load is below the configured threshold and no qualifying activity is reported or estimated. Idle does not prove that every application has stopped working.
GPU activity, a resident model or a compute context alone does not prove AI inference. Short external operations between samples may be missed. History episodes are approximate; reported-session durations describe application-reported activity and can overlap. Neither is exact GPU inference time, billing or token accounting.
Make the workflow yours with a monitored GPU selection, accent color, optional blinking, exact process-name rules, exclusions and local retention settings. Selecting a GPU observes it; it does not route or accelerate workloads. Start with Windows is optional and off by default. If you enable it and save, the Store edition uses Windows' package startup task to open in the tray at sign-in, subject to Windows and administrator settings. Closing the window keeps monitoring in the tray; choose Exit from the tray menu to stop it.
Monitoring works locally without an app account, model download, subscription, advertising or analytics. Recording is on by default with seven-day retention, adjustable from one to 90 days. Disable recording or clear stored history in Settings. The app does not inspect AI prompts, responses, model contents, process command lines or process memory, and does not automatically upload monitoring data. Exports can contain process names, IDs and GPU identifiers; review them before sharing.
REQUIREMENTS AND LIMITATIONS
• Windows desktop, x64 package. A currently supported Windows 11 release is recommended. The package installation floor is Windows 10 build 19041; that technical floor is not a guarantee of compatibility or current Microsoft support for every older Windows release.
• A compatible NVIDIA GPU and installed NVIDIA driver exposing NVML are required for hardware readings. Availability of individual metrics and process attribution depends on the GPU, driver and Windows counters. Missing readings show N/A; no simulated GPU readings are substituted.
• AMD, Intel GPU and NPU monitoring are not supported. CPU-only and remote AI activity are not monitored. This app does not identify all AI workloads, models or Tensor Core use, run inference, tune hardware, or promise performance gains.
• This self-contained Store package includes its .NET runtime. No separate .NET SDK, CUDA toolkit, AI model or local web server is required. No specific RAM or VRAM minimum is asserted; there is no such enforced limit in this release.
Independent application, not affiliated with, sponsored by or endorsed by NVIDIA. NVIDIA and RTX are trademarks of NVIDIA Corporation.