Use cases

Watch a GPU server or Jetson while it trains or serves models

GPU busy time, temperature and memory on the device page, alerts when it overheats or fills its disk, and a shell to check on a job.

For: people running local LLMs, training jobs or edge AI on NVIDIA, AMD or Jetson hardware.

The agent reads the GPU and the system, and you watch from anywhere.GPU box · Jetsonnvidia-smi, sysfsFieldhandrelay.fieldhand.devYoubrowser or sshoutbound :443charts · alerts · shell
The agent reads the GPU and the system, and you watch from anywhere.

What you get

  • GPU busy time from nvidia-smi on NVIDIA cards, from sysfs on AMD and on Jetson, next to CPU, memory, disk and temperature, with 24-hour charts.
  • Alerts when the hottest sensor passes 80 °C or the disk passes 90%: checkpoints fill disks fast.
  • A shell or fh ssh to look at a job, and fh forward box 8888 for Jupyter.
  • The inventory lists the GPU, CPU and USB devices of each box.

Good to know

  • GPU busy time is sampled at most once a minute on NVIDIA, to keep nvidia-smi cheap.
  • It's a health view, not a job scheduler or an experiment tracker.

Start

Create a workspace at fieldhand.dev (the first 5 devices are free, with every feature), click Add devices and run the command on the machine:

curl -fsSL https://api.fieldhand.dev/install.sh | sudo sh -s -- --token fhe_…

Or look around the live demo fleet first.

Something wrong or missing on this page? Tell us.