Trending Model:#1Qwythos-9B-Claude-Mythos-5-1M-GGUFempero-ai⬇1251kTrending Model:#2Unlimited-OCRbaidu⬇758kTrending Model:#3GLM-5.2zai-org⬇176kTrending Model:#4Ornith-1.0-35B-GGUFdeepreinforce-ai⬇285kTrending Model:#5Ornith-1.0-9B-GGUFdeepreinforce-ai⬇255kTrending Model:#6gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUFyuxinlu1⬇314kTrending Model:#7Ornith-1.0-9Bdeepreinforce-ai⬇58kTrending Model:#8DeepSeek-V4-Pro-DSparkdeepseek-ai⬇8kTrending Model:#9Ornith-1.0-35Bdeepreinforce-ai⬇186kTrending Model:#10Qwen-AgentWorld-35B-A3BQwen⬇39kTrending Model:#1Qwythos-9B-Claude-Mythos-5-1M-GGUFempero-ai⬇1251kTrending Model:#2Unlimited-OCRbaidu⬇758kTrending Model:#3GLM-5.2zai-org⬇176kTrending Model:#4Ornith-1.0-35B-GGUFdeepreinforce-ai⬇285kTrending Model:#5Ornith-1.0-9B-GGUFdeepreinforce-ai⬇255kTrending Model:#6gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUFyuxinlu1⬇314kTrending Model:#7Ornith-1.0-9Bdeepreinforce-ai⬇58kTrending Model:#8DeepSeek-V4-Pro-DSparkdeepseek-ai⬇8kTrending Model:#9Ornith-1.0-35Bdeepreinforce-ai⬇186kTrending Model:#10Qwen-AgentWorld-35B-A3BQwen⬇39k

Boxwrench Delivers xdna-top To Track Local AI Chip Activity Easily

By vramkickedin | | 2 min read
Futuristic microchip made up of a frosted glass and matte black brushed metal texture.

The new tool called xdna-top is a terminal system monitor built specifically for AMD Ryzen AI processors. It provides a unified, real-time view of both the neural processing unit and integrated graphics processing unit. Users can see exactly how their hardware handles local artificial intelligence workloads at a glance.

Developer Boxwrench created this software to solve a major blind spot in current monitoring tools. While existing programs show graphics activity, they completely miss the new artificial intelligence chip activity. This project fills that gap by polling the kernel and driver counters to display real hardware context and activity.

Live telemetry for local AI

Key Features
  • Unified live view of both engines.
  • Per-context neural processing unit activity.
  • Json mode for scripts and logging.
  • Snapshot mode for capturing telemetry evidence.
  • Prometheus exporter serving metrics for scraping.
  • Direct driver power state level reporting.
  • Zero daemon and zero root dependency.

This utility is ideal for anyone running local large language models on AMD Strix Halo machines. People can track exactly which processes use the neural chip and verify their inference requests are working. It helps users optimize their system setup by proving hardware activity instead of relying on guesswork.

Project notes and future plans

Currently, the software requires Linux with the proper driver bound and at least Python version 3.11 installed. The developer built the tool with pessimistic fallbacks, meaning it degrades gracefully instead of crashing when a driver node is missing. Future updates plan to move toward direct device probes and support older Ryzen AI generations.

"Born from a practical need: while experimenting with concurrent NPU + iGPU local LLM inference on Strix Halo, "is the NPU actually executing?" turned out to be unanswerable with stock tools." Source: GitHub

Project Links