QUICKSTART / FIRST WORKLOAD
Launch a GPU workload
Create an ephemeral lease against the live node registry. The scheduler resolves hardware constraints, region affinity, spot ceiling, and image compatibility before requesting payment authorization.
1. Configure your client
$ tgpu auth wallet --network mainnet-beta $ tgpu context set region eu-central $ tgpu context set max-spot 0.50
2. Submit a workload specification
apiVersion: tgpu.io/v1alpha2
kind: Workload
metadata:
name: train-mistral-7b
spec:
accelerator: nvidia/rtx4090
count: 1
image: pytorch/pytorch:2.4-cuda12.4
resources:
memory: 48Gi
storage: 200Gi