Gputaskman

Gputaskman is a terminal UI GPU task manager that tracks per-process memory, engine activity, and LLM details.

Gputaskman is a terminal user interface that connects per-process GPU memory and engine activity with loaded runtimes, model files, and large language model buffer breakdowns. It answers what is using GPU memory, such as model weights, KV cache, browsers, or desktop compositors. The software targets local GPU workloads including decision models, vision, speech, LLM inference, training, fine-tuning, graphics, and video. It supports Windows with NVIDIA and AMD GPUs, Linux and WSL, and macOS on Apple Silicon. The tool measures device and process counters while providing estimates for classifications and GGUF-derived breakdowns.

Gputaskman at a glance

Best for: Software engineer
Runs on: Windows, Linux, WSL, macOS, CLI
Pricing: Free
Released: November 2026
Category: AI productivity tool
Website: github.com/shanselman/gputaskman

Key features

Track per-process VRAM and GPU activity
Inspect LLM model files and KV-cache insights
Monitor Windows, Linux, and WSL environments
View engine load, temperature, power, and clocks
Pair unresolved PIDs in container environments
Read amdgpu metrics on native Linux without ROCm

Use it for

Monitor local GPU memory usage by specific processes
Inspect LLM model weights and KV cache breakdowns
Track GPU engine load across containers and WSL
Analyze graphics, video, and AI workload resources

Alternatives to Gputaskman

GPUStack — Unifies AI model serving and GPU compute management across heterogeneous hardware.
RigCheck — Calculates exact model memory limits and file requirements based on developer specifications.
GPUWorks — Manages NVIDIA, AMD, Ascend, PPU, Hygon, and Apple Silicon in one cluster.
Llmspeed — Simulates tokens per second and time-to-first-token using verified benchmarks.
LLM Usage Desktop Widget — Tracks LLM usage limits directly in the system status bar without manual checks.
Dragomand — Provides shared, offline machine translation for any Linux application.
Trim 1 — Runs on one CPU core in milliseconds with no GPU or network calls required.
Klara — It keeps all mobile screen activity private and searchable on-device.
Jantt — Generates structured schedules for AI agents with fewer tokens.
GitHub Stars Manager — Local-first repository organization with AI-powered search and release tracking.
Screenpipe — It provides long-term memory for teams and AI agents while keeping raw history local.
Byakugan Scan — Zero disk writes and under 50ms processing speed on Apple Silicon.

All Gputaskman alternatives