[ThinkOffApp]

Petrus asked for a GPU line on the device cards (room, 2026-11-02 19:15Z). This is the heartbeat half; the card row is ThinkOffApp/groupmind claudemm/device-gpu-row.

What changes

• host-telemetry.js reads the first GPU's load, temperature, SM clock and power from nvidia-smi on Linux and publishes them as gpu_pct, gpu_temp_c, gpu_mhz and gpu_w.
• Same contract as the other fields: a value nvidia-smi prints as [N/A], or an implausible one, is omitted, never zeroed. A host without nvidia-smi publishes none of the four, and macOS never asks.
• Why: the card's TEMP is the hottest board thermal zone. On a GB10 under load that read 84 to 90 °C while the GPU itself was 62 to 72 °C at 95%, so the board number alone cannot say how busy the GPU is.

Checked

• node --test test/*.test.js: 179 passed, 0 failed (6 new tests: full reading, an [N/A] field, idle 0% kept, no nvidia-smi, macOS, implausible values).
• On a GB10 (asus1), the real module returned gpu_pct 95, gpu_temp_c 62, gpu_mhz 2190, gpu_w 23.5.
• Running now on asus1, 2, 3, 5 and 6 (installed by hand, old file kept as a backup); all five heartbeats restarted clean.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-02T19:32:02.422231Z	00a7578	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.