Finally, A Task Manager That Actually Tells You Which AI Ate Your Entire GPU Memory!
While modern operating systems still treat your expensive graphics card like a mysterious black box that just gets hot, a veteran engineer built an open-source tool to finally expose the truth.
When trying to figure out what exactly is consuming gigabytes of graphics memory on a modern desktop, users usually encounter a stone wall of corporate minimalism. Windows offers a performance tab showing little more than temperature and overall load, while Nvidia's built-in tools merely list active processes without explaining why a background script is hoarding video RAM. To get any real answers, administrators have historically had to resort to arcane PowerShell commands just to understand basic resource allocation.
Built by Microsoft engineer and vice-president Scott Hanselman, the open-source utility called gputaskman is designed to bridge this frustrating visibility gap across Windows, Linux, WSL, and macOS. The entire source code was drafted with GitHub Copilot in Python and released under an MIT license, proving that even corporate veterans occasionally need AI assistance to fix their own platform's blind spots. Rather than dumping raw VRAM contents or inventorying game textures, the software focuses strictly on local workloads like computer vision, speech recognition, and large language model inference.
The utility breaks down core metrics per individual process, tracking GPU utilization, temperature, power draw, clock speeds, and the intensity of PCIe bus traffic alongside dedicated encoder and decoder activity. When local artificial intelligence models are active, the interface surfaces specific runtime environments, quantization parameters, layer counts, and exact memory footprints. Future development plans aim to expand support beyond GGUF formats to include ONNX Runtime, PyTorch, and TensorRT, alongside native DRM fdinfo integration for AMD and Intel hardware on Linux.
The era of blindly trusting a generic 95 percent utilization spike without knowing whether it is a rogue browser tab or a local neural network has quietly come to an end, leaving operating system vendors with some awkward questions about basic user transparency.
Source: Neowin
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.