Microsoft VP Builds a Free Tool That Shows Exactly Which App Is Eating Your VRAM
Windows Task Manager tells you that 15 GB of VRAM is in use. It never tells you what is actually using it. GPU Taskman, a new free and open source tool from Microsoft VP Scott Hanselman, exists to answer that exact question, process by process, app by app, and even model by model for local AI workloads.
Hanselman announced the project on X on November 3, 2026, saying he wanted better visibility into where his own GPU memory was going and inviting other developers to contribute. A week later, the tool is already on version 0.3.5, and it is shaping up into the one utility every PC gamer with a VRAM hungry game should know about.
What GPU Taskman Actually Shows You
GPU Taskman is a terminal based interface that pulls together VRAM consumption, GPU accessible system memory, individual process activity, and hardware telemetry in one place. Wccftech reports that it displays GPU utilization, VRAM usage, temperature, power consumption, clock speeds, memory activity, and PCIe statistics, alongside a breakdown of every process consuming GPU resources.
The key difference from the tools you already have is the per process view. Instead of one big VRAM number, you see which app is responsible for how much of it. The tool can tell desktop compositing apart from a web browser, a game, and an AI application, so you can finally find out whether it is your game eating VRAM or the 40 browser tabs sitting behind it.
That matters because VRAM pressure is one of the quietest causes of bad performance on PC. As Wccftech notes, excessive VRAM pressure can contribute to stuttering when applications start leaning on slower system memory. If your game hitches every few seconds at high settings, the first thing worth checking is what else is holding VRAM hostage. Our guide to how to fix game stuttering on PC covers the usual suspects, and GPU Taskman now gives you a way to confirm the culprit instead of guessing.
The AI Angle Is the Real Surprise
The most interesting feature has nothing to do with games. GPU Taskman can identify supported local AI models and report their architecture, quantization, parameter counts, model weights, and estimated KV cache requirements. It recognizes runtimes including llama.cpp, Ollama, and LM Studio, although the level of detail varies by runtime and by the information available.
In practice, that means the tool can split an AI app’s VRAM footprint into useful pieces instead of one opaque number. Rather than simply reporting that an AI application occupies several gigabytes, it can help distinguish memory allocated to model weights from memory used by the KV cache and compute buffers. If you run local AI models on the same GPU you game on, that breakdown is genuinely useful.
There are honest limits here, and Hanselman states them openly. Some model specific values are estimates rather than directly measured GPU allocations, because the tool combines operating system GPU counters with process metadata, model files, runtime logs, and local APIs. The FPS Review adds that identifying individual models requires supporting evidence, and that a PyTorch or CUDA file alone cannot prove a model is resident on the GPU. The tool is upfront about what it knows and what it is guessing.
What It Cannot Tell You
GPU Taskman is not magic, and its creator is clear about the boundaries. Hanselman cautions that GPU accessible system RAM usage does not necessarily indicate VRAM overflow, so a big system RAM number is not automatic proof that your card is out of memory. The tool also cannot identify individual game textures or determine precisely which resources have been evicted from VRAM.
That honesty is part of the appeal. The tool tells you what can be known about GPU memory and labels the rest as estimates. If you want the basics first, our guide on how to check your GPU temperature while gaming covers the standard monitoring tools, and GPU Taskman sits naturally next to them as the deeper, process level view.
Platform Support and the Latest Release
GPU Taskman supports NVIDIA and AMD GPUs on Windows 10 and 11, alongside Linux, WSL, and Apple Silicon macOS systems, with capabilities varying by platform and hardware. The latest v0.3.5 release, published on November 10, adds Intel GPU support on Linux, plus CPU and system RAM monitoring and additional options for recording GPU statistics.
| Platform | GPU Support | Notes |
|---|---|---|
| Windows 10 and 11 | NVIDIA and AMD | Portable builds available, no Python install needed |
| Linux | NVIDIA, AMD, and Intel (v0.3.5) | Intel GPU support added in the November 10 release |
| WSL | NVIDIA and AMD | Capabilities vary by setup |
| macOS (Apple Silicon) | Apple Silicon | Capabilities vary by platform |
The utility is distributed under the MIT license and can be downloaded from its GitHub Releases page. Portable Windows builds are available without requiring a separate Python installation, and users can also install it through PyPI. Development is moving fast, with new releases landing within days of each other.
Why PC Gamers Should Care
VRAM is becoming the most argued about spec in PC gaming. Games keep asking for more of it, 8 GB cards keep getting squeezed at high settings, and driver level tools rarely explain where the memory actually went. A free tool that names the process behind every gigabyte is a practical addition to any gamer’s toolkit, right next to a temperature monitor and a frame rate overlay.
The use cases are simple. A game stutters at ultra textures, and you want to know if the game itself is the problem or if your browser and streaming app are eating 3 GB in the background. A local AI model is running while you play, and you want to know how much VRAM its weights and KV cache are holding. A new driver lands, and you want to see whether VRAM behavior actually changed. GPU Taskman answers all three without a paid license or a heavyweight install.
It is still experimental software, and some monitoring capabilities need further validation across different hardware configurations. Treat the numbers as a guide, not gospel, especially the AI model estimates. But for a free, open source tool built in the open by a Microsoft VP, the starting point is already strong.
There is even a rumor, noted by The FPS Review, that the tool could eventually be integrated into Windows itself. Nothing is confirmed, and for now the GitHub release is the only way to get it. If that integration ever happens, it would be the clearest sign yet that per process VRAM visibility belongs in the operating system, not just in third party tools.
If you are tuning a rig for demanding games, pair GPU Taskman with our guide to the best PC settings for maximum FPS. Knowing which app is eating your VRAM is only half the job. The other half is turning that knowledge into settings that actually run smoothly.
How to Get GPU Taskman
GPU Taskman is free, open source, and available now from its GitHub Releases page, with portable Windows builds that need no Python installation and a PyPI install option for everyone else. Version 0.3.5 is the latest release as of November 10, 2026, and development is progressing rapidly, so expect more features and wider hardware support in the coming weeks.
For a tool that started as one developer scratching his own itch, it has arrived at exactly the right moment. VRAM is scarce, games are hungry, and local AI is now sharing your GPU with everything else. Finally, there is a free way to see who is eating what.
Source: Wccftech, The FPS Review
