The gputaskman VRAM AI monitor is the new open-source tool presented by Scott Hanselman, VP and Member of Technical Staff at Microsoft and GitHub. Published on GitHub under the MIT license, it offers detailed per-process visibility into VRAM usage, GPU activity, and specific insights for AI and LLM workloads.
The project came from a practical need Hanselman shared on X on November 3, 2026: “There's a lot going on in my GPU. I'm making a GPU Taskman to get insights to where my GPU memory is going.” Windows Task Manager shows total VRAM used but does not break down what is consuming it — model weights, KV cache, the browser, or the desktop compositor. gputaskman fills that gap clearly and accessibly.
The terminal interface (htop-style) displays, per process:
- GPU percentage usage and memory activity
- Consumed VRAM, spill, and paged memory
- Temperature, power draw, clocks (SM, memory, video)
- Encoder/decoder activity and PCIe traffic
- For AI workloads: model name, weights, KV cache size, and parameter count (e.g., in processes like llama-server.exe)
The tool works on Windows (NVIDIA and AMD), Linux/WSL, and macOS, with portable builds available on GitHub. It is experimental and does not arbitrarily inspect GPU memory — it uses heuristics and system APIs. Hanselman has indicated interest in expanding support (including Windows on Arm) and suggested similar capabilities could eventually reach native Windows tools.
For anyone running local models, monitoring VRAM on gaming PCs, or simply wanting to know why the GPU is busy, the gputaskman VRAM AI monitor offers transparency that standard tools do not. The code is at https://github.com/shanselman/gputaskman.
Why it matters
With the growth of local AI and models that consume significant VRAM, precise monitoring tools become essential. gputaskman helps identify bottlenecks, optimize hardware use, and understand the real impact of AI applications on the system.
Sources
- Tom’s Hardware (November 8, 2026)
- Scott Hanselman's GitHub repository
- Original X post (@shanselman, November 3, 2026)
This article was produced with the assistance of artificial intelligence and reviewed for editorial accuracy.
Por Grok
