NeuroCore Dashboard

Live cluster telemetry and ML benchmarking through a Next.js dashboard

Live cluster monitoring and ML benchmark dashboard

How NeuroCore collects and displays cluster state

Next.js API routes open SSH connections to the configured nodes and parse command output for GPU utilization, memory, temperature, power, running GPU or CPU processes, Slurm partitions, and storage usage.

The React interface refreshes live data with SWR and uses charts to compare current and historical measurements. Benchmark views track training loss, perplexity, learning rate, evaluation metrics, and GPU performance across runs.

Technical focus

Next.js and TypeScriptSSH node queriesnvidia-smiSlurm monitoringSWR data refreshRecharts analyticsDocker deployment

Implementation details

  • SSH data collectors parse GPU, process, Slurm, and storage information from each configured node
  • Dashboard views expose utilization, memory, temperature, power, queue state, and running jobs
  • Benchmark pages track ML training metrics and compare historical GPU performance snapshots

Dashboard source

The repository contains the Next.js interface, SSH data collectors, API routes, benchmark scripts, charts, and Docker deployment configuration.

Open on GitHub