NVTopal -- GPU Monitor

NVTopal is a terminal-based NVIDIA GPU monitor inspired by htop and nvtop. It parses nvidia-smi output to render real-time metrics for every GPU in your system, with a keyboard-driven interface.

Key Features

  • Multi-GPU view -- All NVIDIA GPUs (0-N devices) in a single screen.
  • Real-time graphs -- Sparklines for utilization, memory, temperature, power, clock speeds, and fan speed.
  • GPU load averages -- 1-, 5-, and 15-minute load averages, the same shape as Linux's CPU load.
  • Per-model GPU shader specs -- Built-in table of SM count, ROPs, and TMUs for accurate per-model reporting.
  • Configurable -- Layout, refresh rates, sort order, and history depth via config.yaml.

Requirements

  • nvidia-smi on PATH (Linux, macOS, or Windows).
  • One or more NVIDIA GPUs visible to the driver.

Content for this section has not been authored yet.

Who Uses NVTopal?

NVTopal is for anyone who wants a terminal-resident, real-time view of NVIDIA GPU activity — the way htop shows CPU.

Back to NVTopal


Running compute workloads

You're training a model, running local inference, or driving any sustained GPU workload, and you want to see whether utilization, memory, temperature, or clocks are drifting across the run. NVTopal keeps a live sparkline history per metric in a single terminal pane.

Multi-GPU machines

You have 2-8+ GPUs and want one screen that shows all of them at once, with totals and per-device detail. Sort by utilization, memory, or power to see which device is hot.


Where NVTopal Doesn't Fit

  • AMD or Intel GPUs — NVTopal reads nvidia-smi and is NVIDIA-only.
  • Historical retention and alerting — NVTopal is a live terminal tool. For long-term trending or alerts, use Prometheus + a GPU exporter.
  • Multi-host visibility — NVTopal monitors the local machine. For fleet-wide GPU visibility, use a centralized metrics platform.

Back to NVTopal

Content for this section has not been authored yet.

Content for this section has not been authored yet.

Content for this section has not been authored yet.