Skip to main content

Launching GPU Instances

Deploy on-demand GPU instances in minutes.

Prerequisites​

  • Wallet funded with at least 1 hour of instance cost
  • At least one SSH key added to your account

Not ready? Complete Getting Started first.

Browse GPU Marketplace​

The Home page displays available GPU configurations in real-time.

GPU Listing Information​

SpecificationDescription
GPU ModelType and count (e.g., A100, H100, RTX 4090)
VRAMGPU memory per card
vCPUNumber of CPU cores
RAMSystem memory
StorageNVMe instance storage
RegionDatacenter location
PriceHourly rate in EUR (€)
StatusAvailable / Coming soon

Filter options: Region, GPU type, availability

Note: Availability updates in real-time.

Select Configuration​

Click on a GPU card to view detailed specifications and launch configuration.

Key information displayed:

  • Complete hardware specs
  • Pricing (hourly, daily, monthly estimates)
  • Region and cloud provider
  • Operating system options

Configure Instance​

Required Settings​

Instance Name:

  • 1-50 characters
  • Use descriptive names (e.g., ml-training-01, render-job-42)

Region:

  • Pre-selected based on GPU availability
  • Cannot be changed

SSH Key:

  • Select from your added keys
  • Required for instance access

Operating System:

  • Choose from available images
  • All images include CUDA drivers and GPU tools
  • Format: Ubuntu version + CUDA version

Optional Settings​

Tags:

  • Add labels for organization (e.g., production, ml-training, team-alpha)
  • Useful for filtering and cost tracking

Environment Variables:

  • Set key-value pairs for your instance
  • Example: API_KEY=your_key, MODEL_PATH=/mnt/volume/models

Startup Script:

  • Bash script executed on first boot
  • Use for installing software, downloading data, configuring services
  • Runs as root user

Example:

#!/bin/bash
apt-get update
pip3 install torch torchvision transformers
cd /root
git clone https://github.com/your-repo/project.git

Storage Volume:

  • Attach existing volume or create new volume
  • Volumes must be in same cloud region as instance
  • Choose size (50 GB to 10,000 GB)
  • Additional hourly cost applies

Launch​

  1. Review pricing estimate
  2. Verify wallet balance is sufficient (minimum 1 hour cost)
  3. Click "Launch Instance"

The system checks availability in real-time. If available, the first hour is charged and instance creation begins.

Instance Status​

After launch, instances progress through these states:

  • creating: Infrastructure provisioning (1-2 minutes)
  • pending_provider: Cloud provider setup (2-5 minutes)
  • pending: Final configuration (1-2 minutes)
  • active: Running and accessible
  • failed: Launch failed (first hour refunded)
  • suspended: Insufficient wallet balance
  • deleting: Terminating (1-2 minutes)
  • deleted: Removed

Total launch time: 5-10 minutes

Connect via SSH​

Once status is active, connect using:

ssh -i /path/to/private_key.pem root@<IP_ADDRESS>

Connection details (IP address, SSH port, SSH user) are available on the Resources page.

Verify GPU access:

nvidia-smi

Common Issues​

GPU Unavailable:

  • Try different region or GPU type
  • Availability changes in real-time

Insufficient Balance:

  • Add funds to wallet
  • Minimum required: 1 hour of total cost (instance + volume)

Cannot Connect:

  • Wait for status to be active
  • Verify correct IP address and private key
  • Check key permissions: chmod 600 key.pem
  • Ensure status is active, not creating or pending

Instance Stuck in Creating:

  • Normal duration: 5-10 minutes
  • If exceeds 15 minutes, contact support
  • Failed instances automatically refund first hour

Next Steps​