Run Datature Training on a Self-Hosted GPU Runner
Train on Your Own Hardware, Managed by Datature
Not every team wants to run training on cloud GPUs. Some organizations have compliance requirements that keep data on-premise. Others have underutilized GPU servers sitting in their own data centers. The self-hosted GPU runner lets you use Datature Nexus to orchestrate training while the actual computation happens on your hardware.
What This Tutorial Covers
Datature walks through setting up the self-hosted runner:
- Installing the runner agent on your GPU machine
- Connecting the runner to your Datature workspace
- Selecting the self-hosted runner when launching a training workflow
- Monitoring the training run from the Nexus dashboard
Setup takes about three minutes. After installation, the runner stays connected and is available for future training jobs without reconfiguration.
When Self-Hosted Makes Sense
This option is a good fit for teams in regulated industries (healthcare, defense, financial services) where training data cannot leave the network. It also works for teams with existing NVIDIA GPU infrastructure that want Datature's workflow management without paying for additional cloud compute. The runner supports standard NVIDIA GPUs and handles data transfer, checkpointing, and artifact storage automatically.
For teams without dedicated GPUs, Datature also offers cloud-based training. See How to Interpret Training Graphs for tips on monitoring your runs.
