Syllo/nvtop

GPU & Accelerator process monitoring for AMD, Apple, Huawei, Intel, NVIDIA and Qualcomm

What it solves

NVTOP provides a real-time, interactive task monitor for GPUs and AI accelerators, similar to how htop works for CPUs. It solves the problem of needing a unified, visual way to monitor hardware utilization, temperature, and power consumption across a wide variety of hardware vendors in a single terminal interface.

How it works

NVTOP queries hardware information through vendor-specific libraries and drivers (such as NVML for NVIDIA, libdrm for AMD/Intel, and various specialized libraries for accelerators like Tenstorrent or Huawei Ascend). It then uses the ncurses library to render a visual dashboard in the terminal, displaying metrics like GPU/memory utilization, clocks, and PCIe throughput.

Who it’s for

It is designed for developers, data scientists, and system administrators who need to monitor the performance and health of GPUs and AI accelerators during model training, inference, or general compute tasks.

Highlights

  • Broad Hardware Support: Compatible with NVIDIA, AMD, Intel, Apple (M1/M2), Huawei Ascend, Tenstorrent, Rockchip, MetaX, Enflame, and Iluvatar CoreX.
  • Interactive Interface: Includes a setup utility (F2) to customize which metrics are plotted.
  • Detailed Metrics: Tracks GPU and memory utilization, temperature, power, clocks, and PCIe RX/TX load.
  • Process Monitoring: Displays a list of processes currently utilizing the GPU hardware.

Related

  • Project
    getagentseal/codeburn

    CodeBurn is a free, open‑source Node.js utility that locally reads session logs from dozens of AI coding tools (Claude, Cursor, Codex, Gemini, etc.) and shows you exactly how many tokens and dollars each model, project, and task has consumed. It provides a terminal UI, a web dashboard, and a menubar/tray app, can detect waste patterns (re‑reads, low read‑to‑edit ratios, unused agents, etc.), suggest and apply fixes, enforce budget caps, compare model efficiency, and even correlate spend with git commits. All processing stays on your machine; no API keys or data are sent elsewhere.

  • Project
    kedro-org/kedro

    A Python framework for creating production-ready data engineering and data science pipelines using software engineering best practices.

  • Project
    huggingface/tokenizers

    A high-performance Rust-based library for training and implementing the most widely used text tokenization algorithms for AI models.

  • Project
    optiscaler/OptiScaler

    A middleware tool that lets users replace the temporal upscalers and frame generation technologies in games that already support DLSS, FSR, or XeSS.