Ubuntu environments
Available Ubuntu configurations for Spheron GPU instances.
Available versions
| Version | Support until | Best for |
|---|---|---|
| Ubuntu 20.04 LTS | 2025 | Legacy applications, older dependencies |
| Ubuntu 22.04 LTS | 2027 | Production workloads (most stable) |
| Ubuntu 24.04 LTS | 2029 | Latest features, experimental projects |
Configuration options
Base images
- Base / LTS Base: minimal Ubuntu installation
- + NVIDIA 550/570: NVIDIA drivers pre-installed
- + CUDA X.X: CUDA toolkit included
- + Docker: Docker pre-installed for containerized workflows
- Shade OS: optimized lightweight version for maximum GPU performance
Pre-configured ML environments (Ubuntu 24.04)
| Environment | Includes | Best for |
|---|---|---|
| ML Everything | PyTorch, TensorFlow, JAX | Multi-framework experimentation |
| ML PyTorch | PyTorch optimized | LLM training, computer vision |
| ML TensorFlow | TensorFlow optimized | Production ML, enterprise |
Selection guide
| Use case | Recommended environment | Why |
|---|---|---|
| Getting started | Ubuntu 24.04 ML Everything | All frameworks pre-installed |
| LLM training | Ubuntu 24.04 ML PyTorch | PyTorch with CUDA pre-installed |
| TensorFlow | Ubuntu 24.04 ML TensorFlow | TensorFlow with CUDA pre-installed |
| Production | Ubuntu 22.04 + CUDA 12.8 + Docker | Stable, containerized |
| Research | Ubuntu 24.04 + CUDA 13.0 Open | Latest features |
| Legacy apps | Ubuntu 20.04 LTS | Older dependency support |
| Max performance | Ubuntu 22.04 (Shade OS) | Optimized, minimal overhead |
Docker vs non-Docker
Without Docker:- Direct GPU access
- Simpler setup
- Single-purpose instances
- Good for learning and simple projects
- Containerized workflows
- Dependency isolation
- Multi-project instances
- Good for production and complex setups
Deploy
- Go to app.spheron.ai → Deploy
- Select a GPU
- Choose an Ubuntu environment from the OS dropdown
- Click Deploy (ready in 30 to 60 seconds)
Verify installation
After deployment, connect and verify:
# Connect
ssh root@<your-instance-ip>
# Check OS version
cat /etc/os-release
# Check CUDA (if applicable)
nvcc --version
# Check GPU
nvidia-smi
# Check Docker (if applicable)
docker --versionFrequently asked questions
Q: What does LTS mean?
A: Long Term Support: 5 years of security updates and bug fixes.
Q: Can I change environments after deployment?
A: No. Deploy a new instance with the desired environment.
Q: Do I need Docker?
A: Not for simple projects. Use Docker for complex dependencies or multi-project instances.
Q: Which CUDA version should I use?
A: See CUDA and NVIDIA Drivers for a full guide including framework compatibility tables and version recommendations.
Q: Can I install multiple CUDA versions?
A: Not recommended. Select the correct version during deployment.
Q: Ubuntu 22.04 or 24.04?
A: 22.04 for production stability. 24.04 for latest features.
Q: What is Shade OS?
A: An optimized Ubuntu variant with minimal overhead for maximum GPU performance.
What's next
- CUDA and NVIDIA Drivers: CUDA versions, driver details, and how to choose
- Getting Started: Deploy your first instance
- SSH Connection: SSH setup guide
- TensorFlow: TensorFlow environment setup
- Jupyter: Jupyter Notebook setup