Summary
Built a Glance custom-api widget that surfaces a live DGX Spark cluster summary — GPU temperature/utilization, VRAM, LLM generation tok/s, KV-cache usage, and uptime for the head node plus one worker — right on a self-hosted Glance dashboard. Wanted to flag it here in case it's worth linking from the sparkDash docs as an integration.
What it does
- Fetches
/api/sparks/:id/metrics for two units concurrently using Glance subrequests.
- Renders both nodes in one card (online dot, model id, GPU/VRAM/tok-s/ctx%/uptime), with the title linking through to the full dashboard.
- Configurable via
${SPARKDASH_HEAD_URL} / ${SPARKDASH_NODE2_URL} / ${SPARKDASH_DASHBOARD_URL} env vars.
Links
Suggestion
If this seems useful, a short "Integrate with Glance" section in the README pointing to the widget would make the integration discoverable for the self-hosted/Glance crowd. Happy to open a docs PR for that too if you'd like.
Summary
Built a Glance
custom-apiwidget that surfaces a live DGX Spark cluster summary — GPU temperature/utilization, VRAM, LLM generation tok/s, KV-cache usage, and uptime for the head node plus one worker — right on a self-hosted Glance dashboard. Wanted to flag it here in case it's worth linking from the sparkDash docs as an integration.What it does
/api/sparks/:id/metricsfor two units concurrently using Glance subrequests.${SPARKDASH_HEAD_URL}/${SPARKDASH_NODE2_URL}/${SPARKDASH_DASHBOARD_URL}env vars.Links
Suggestion
If this seems useful, a short "Integrate with Glance" section in the README pointing to the widget would make the integration discoverable for the self-hosted/Glance crowd. Happy to open a docs PR for that too if you'd like.