Analysis updated 2026-05-18
Watch real-time GPU and NPU busy percentage while running a local AI model on Strix Halo hardware.
Export a single telemetry reading as JSON for use in scripts or reports.
Record a timed session and run assertions in CI to prove a workload actually used the NPU.
| boxwrench/xdna-top | 0whitedev/detranspiler | 2951461586/mulerun-pool | |
|---|---|---|---|
| Stars | 21 | 21 | 21 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | hard | moderate |
| Complexity | 3/5 | 4/5 | 3/5 |
| Audience | ops devops | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Needs Python 3.11+, the amdxdna driver, and xrt-smi on the system path for full NPU pane support.
xdna-top is a terminal-based monitoring tool for Linux machines running AMD Ryzen AI processors, specifically the Strix Halo family. It shows, side by side, how busy the integrated GPU and the NPU (Neural Processing Unit, the dedicated AI chip on these processors) are at any given moment. It was built because existing tools like amd-smi could not surface this information on Strix Halo hardware, leaving anyone running local AI models with no way to tell whether the NPU was actually doing work. The tool runs in your terminal at a 5 Hz refresh rate and displays two panes. The integrated GPU pane shows a live busy percentage and power draw, pulled directly from Linux kernel counters. The NPU pane lists active hardware contexts by process ID, along with submission and completion counters that let you see whether work is queued or in flight. The tool derives an activity state from counter changes rather than guessing at a utilization percentage, because the underlying driver does not yet expose one directly. Beyond the live dashboard, xdna-top includes several command-line modes for scripting and evidence collection. You can export a single telemetry reading as JSON, record a stream of readings over a timed window, annotate the recording with named markers, and then run assertions against it to confirm the NPU was active within a specific window. This makes it possible to use the tool in automated tests or CI pipelines to prove that a workload actually used the NPU. Installation requires Python 3.11 or newer, the amdxdna driver loaded on Linux, and xrt-smi on the system path for the NPU pane. The integrated GPU pane works without xrt-smi. If either dependency is missing, the relevant pane flags itself as degraded rather than crashing the whole tool.
A terminal dashboard for Linux that shows live GPU and NPU utilization on AMD Ryzen AI Strix Halo processors, filling a gap left by existing tools.
Mainly Python. The stack also includes Python, amdxdna, xrt-smi.
No license information is given in the explanation.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly ops devops.
This repo across BitVibe Labs
Verify against the repo before relying on details.