Microsoft vice president Scott Hanselman has revealed an open-source project called gputaskman, a GPU-focused task manager that shows per-process GPU usage and VRAM consumption, including information about AI and large language model tasks.
The tool, hosted on GitHub under a permissive MIT license, works on Windows, Linux, and macOS. It displays per-process GPU usage, VRAM consumption, active runtimes, and additional statistics, including specific information for AI and large language model tasks.
The project emerged from Hanselman's own need to track GPU memory usage. “There's a lot going on in my GPU,” he wrote on social media, adding that he was making the tool “to get insights to where my GPU memory is going.”
Screenshots shared by Hanselman show the tool's terminal-like interface in action on his RTX 4080S system, with the desktop windows manager listed as the most VRAM-hungry process. When an AI application like llama-server.exe is selected, the tool breaks out details such as the model being used (Qwen3-0.6B), the memory split between weights and KV cache, and the number of parameters.
Tom's Hardware news editor Mark Tyson tested the portable Windows download of the latest v0.3.0 release, noting that live-streaming charts for GPU load, VRAM use, and temperature were more prominent in his version of the display than in Hanselman's examples.
In comments on social media, Hanselman hinted that expanded GPU monitoring capabilities could become standard features in Windows.
Analysis
Why This Matters
- Developers and AI enthusiasts can now see exactly which processes and models consume GPU memory, aiding debugging and resource management.
- The tool's open-source nature and cross-platform support fill a gap left by proprietary operating system utilities that lack such granular AI-focused data.
- Hanselman's hint at broader GPU monitoring in Windows suggests the tool could preview future OS features.
Background
GPU task managers built into operating systems often show only basic usage, such as overall memory load, without breaking down which specific application or AI model is responsible. Third-party tools like GPU-Z or Nvidia's own utilities provide some detail, but none have tied it to per-process AI model information in the way gputaskman does.
Key Perspectives
Scott Hanselman (Microsoft VP): developed the tool because existing tools did not give him enough insight into what was consuming GPU memory, particularly for AI workloads.
Mark Tyson (Tom's Hardware news editor): notes the tool's live charts were prominent in his test, and that Hanselman's screenshots might be from an earlier version.
What to Watch
- Whether Microsoft integrates similar per-process VRAM and AI task tracking into a future Windows update, as Hanselman hinted.
- Community adoption of the tool on Linux and macOS, where GPU monitoring tools are less common than on Windows.
- Updates to gputaskman that may add support for more GPU vendors or deeper AI model insights.
Checked
12 claims in this story checked against 1 source: 12 supported.
Show each claim
- SupportedThe headline says Microsoft vice president Scott Hanselman revealed an open-source tool for showing what is using GPU memory.Latest from Tom's Hardware: “Microsoft VP Scott Hanselman has revealed a new open-source GitHub project called gputaskman.”
- SupportedGputaskman works on Windows, Linux and macOS and shows per-process GPU usage, VRAM consumption, active runtimes and AI/LLM task information.Latest from Tom's Hardware: “This new tool for Windows, Linux, and macOS can show per-process GPU usage, VRAM consumption, active runtimes, and many more stats. It also breaks out AI/LLM task-specific information.”
- SupportedHanselman is a Microsoft vice president.Latest from Tom's Hardware: “Microsoft VP Scott Hanselman has revealed a new open-source GitHub project called gputaskman.”
- SupportedGputaskman is hosted on GitHub under a permissive MIT license.Latest from Tom's Hardware: “Hanselman is sharing his “experimental software” tool under the permissive MIT license.”
- SupportedThe project arose from Hanselman's desire to track where his GPU memory was going.Latest from Tom's Hardware: “On social media, Hanselman snappily makes clear the technical itch that gputaskman was designed to scratch.”
- SupportedHanselman wrote, “There's a lot going on in my GPU.”Latest from Tom's Hardware: ““There's a lot going on in my GPU,” wrote the prominent public face of Microsoft's developer ecosystem.”
- SupportedHanselman said he was making the tool “to get insights to where my GPU memory is going.”Latest from Tom's Hardware: ““I'm making a GPU Taskman to get insights to where my GPU memory is going.””
- SupportedHanselman's screenshots show the desktop windows manager as the most VRAM-hungry app on his RTX 4080S system.Latest from Tom's Hardware: “Hanselman’s tweet shows that the most VRAM-hungry app on his RTX 4080S system is the desktop windows manager.”
- SupportedSelecting llama-server.exe shows Qwen3-0.6B, a breakdown of weights and KV cache, and the number of parameters.Latest from Tom's Hardware: “The bottom of the window breakdown reveals that Qwen3-0.6B is being used, and there’s a breakdown of its weights, the KV cache, the number of parameters, and more.”
- SupportedTom's Hardware news editor Mark Tyson tested the portable Windows download of the latest v0.3.0 release.Latest from Tom's Hardware: “As the Microsoft developer has made portable Windows downloads of the latest v0.3.0 release of gputaskman, I couldn’t resist giving it a spin.”
- SupportedTyson noted that the live-streaming GPU-load, VRAM-use and temperature charts were more prominent in his display than in Hanselman's examples.Latest from Tom's Hardware: “I note the live-streaming charts showing GPU load, VRAM use, temperature, and so on are more prominent in my screenshot than in Hanselman’s examples.”
- SupportedHanselman hinted in social media comments that expanded GPU monitoring capabilities could become standard in Windows.Latest from Tom's Hardware: “Finally, in social media comments, he hints that expanded GPU monitoring capabilities could be coming to Windows as standard.”
