gputaskman hits 3.5k GitHub stars: a terminal htop that maps per-process VRAM and LLM KV-cache
alex_verem · x · 2026-11-17
The open-source project gputaskman (shanselman/gputaskman, MIT license) has reached 3.5k stars on GitHub.
- An "htop for what's inside your GPU, not just how full it is"
- Supports Windows (NVIDIA/AMD telemetry, Intel WDDM activity), Linux & WSL (NVIDIA everywhere, AMD/Intel native Linux, plus Huawei Ascend on Linux), and macOS on Apple Silicon
- Terminal UI connects per-process GPU memory and engine activity with loaded runtimes, model files and LLM buffer breakdowns
- Answers "Task Manager says 15GB used — by what?": model weights, KV cache, your browser, or the desktop compositor
- Scope is local GPU workloads (LLM inference, fine-tuning, graphics, video); CPU/NPU-only and cloud are out of scope
Related event: Open-Source gputaskman Gains Traction as htop for GPUs(2 posts)→
More from Infra
- Microcenter posts deepest GPU discounts in months: RTX PRO 6000 drops $4,000, still above MSRP — OvertaxedOne · 2026-11-17
- Cloud outages are a dependency problem, not a Microsoft problem — DavidLinthicum · 2026-11-17
- Intel shares first Crescent Island D270 benchmarks: 29% lower AI latency than RTX PRO 6000 Blackwell — pulse77 · 2026-11-17
- Is remote desktop GPU access a practical alternative to high-end laptops? — robertwellesley · 2026-11-17
- gputaskman: an open-source htop for your GPU that breaks down VRAM by process and LLM KV-cache — alex_verem · 2026-11-17
- Megaspeed Got $800M in Nvidia Chips After No-Sell Listing, Faces Criminal Probe — GaryMarcus · 2026-11-17