Getting Started

Deploy your first GPU instance on Devscale in under 10 minutes. This guide walks through account creation, CLI setup, instance provisioning, and connecting to your server.

1.Create your Devscale account

Free trial credits are available for new accounts so you can test GPU and CPU instances before committing.

2.Install the Devscale CLI

Install the CLI globally via npm. It provides a fast way to create, manage, and SSH into instances.

# npm install -g @devscale/cli

# devscale login

This opens a browser where you authenticate with your Devscale credentials. The CLI stores a session token locally.

3.Create your first GPU instance

Provision a bare-metal GPU instance using the dashboard or CLI. For quick experimentation, an NVIDIA L40S spot instance is the most cost-effective option.

# devscale instances create --model L40S --count 1 --region us-west

The instance will be provisioned in under 2 minutes. You'll receive an SSH IP and key pair automatically.

4.Connect and start building

SSH into your instance and set up your ML environment. Every instance ships with Ubuntu 22.04, CUDA 12.x, and Python pre-installed.

# devscale ssh my-instance-01

$ nvidia-smi

5.Store and retrieve data

Use Devscale Object Storage for datasets, checkpoints, and model artifacts. It's S3-compatible so standard tooling works out of the box.

# devscale storage upload ./checkpoints s3://my-bucket/

Object storage is $0.01/GB/month with zero egress fees — ideal for large training datasets.

API examples

Devscale integrates with standard tools. Use the Python SDK for storage, or the REST API for full platform control.

Python SDK (via boto3)

import boto3

s3 = boto3.client('s3',
    endpoint_url='https://s3.devscale.online',
    aws_access_key_id='YOUR_ACCESS_KEY',
    aws_secret_access_key='YOUR_SECRET_KEY',
)

# Upload a file
s3.upload_file('model.pt', 'my-bucket', 'checkpoints/model.pt')
print('Upload complete')

REST API (curl)

curl -X POST https://api.devscale.online/v1/instances \
  -H "Authorization: Bearer $DEVSCALE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "L40S",
    "count": 1,
    "region": "us-west"
  }'

What's next?

Now that your first instance is running, explore more of the platform.