DOCUMENTATION

Build with Nile GPU.

Everything you need to deploy GPU clusters and run AI workloads.

Quickstart

Get a GPU cluster running in under 60 seconds.

1. Get your API key

Sign up at nilegpu.com and retrieve your API key from the dashboard.

export NILE_API_KEY=nk_live_your_key_here

2. Install the SDK

pip install nile-compute

3. Launch a cluster

import nile

client = nile.Client(api_key="nk_live_...")

cluster = client.clusters.create(
    gpu="h100",
    count=8,
    region="addis-1",
    image="pytorch-2.4-cuda12.4"
)

print(f"Status: {cluster.status}")
print(f"SSH:    {cluster.ssh_command}")
print(f"Cost:   ${cluster.cost_per_hour}/hr")

4. Connect and train

ssh ubuntu@203.0.113.42
python -c "import torch; print(torch.cuda.device_count())"  # 8

5. Terminate when done

result = client.clusters.terminate(cluster.id)
print(f"Runtime: {result['runtime_hours']:.2f} hours")
print(f"Cost:    ${result['total_cost']:.2f}")