GPU environments, ready to use.
Launch PyTorch, ComfyUI and other configured environments for training, notebooks and model development.
Managed containers with per-minute compute billing.
Template catalog
Choose your starting environment.
Each template includes a configured software stack for its workload.
PyTorch
Tensor computation, autograd and GPU acceleration in a Pythonic API.
TensorFlow
End-to-end machine learning platform from Google.
FastAI
High-level PyTorch wrapper for rapid training, vision to NLP.
Axolotl
Fine-tune open LLMs like Llama and Mistral from a YAML config.
ComfyUI
Node-based workflow UI for Stable Diffusion pipelines.
Automatic1111
The Stable Diffusion web UI, with its extension ecosystem.
Fooocus
Text-to-image with minimal prompt engineering.
Kohya
LoRA, DreamBooth and QLoRA fine-tuning for Stable Diffusion.
Ollama
Run open LLMs with a simple CLI and API; bring your own UI.
Development workflow
Bring your code and data.
Use the tools that fit your work without configuring a GPU driver first.
Work in the browser or over SSH
Use JupyterLab, browser-based VS Code or SSH to access your environment and its files.
Install project dependencies
Add packages to the managed environment for your project. Choose a VM when you need your own kernel or Docker daemon.
Explore GPU VMs↗Expose an application
Open an HTTP port for a Gradio demo, FastAPI service or other app on your running instance. For automatic worker scaling, use Serverless.
Explore Serverless↗GPU configurations
Choose compute for your workload.
Match GPU memory and quantity to the model, batch size and framework you use.
USD rates per GPU-hour, billed by the minute; available configurations depend on region and capacity.
- NVIDIA H200 SXM
- GPUs per template
- 1 · 2 · 4 · 8
- Starting rate
- $3.99/hr
- NVIDIA H100 SXM
- GPUs per template
- 1 · 2 · 4 · 8
- Starting rate
- $2.69/hr
- NVIDIA RTX Pro 6000
- GPUs per template
- 1 · 2 · 4 · 8
- Starting rate
- $1.89/hr
- NVIDIA A100-80GB
- GPUs per template
- 1 · 2 · 4
- Starting rate
- $1.49/hr
- NVIDIA A100
- GPUs per template
- 1 · 2 · 4
- Starting rate
- $0.89/hr
- NVIDIA A30
- GPUs per template
- 1 · 2 · 4
- Starting rate
- $0.41/hr
- NVIDIA L4
- GPUs per template
- 1 · 2 · 4 · 8
- Starting rate
- $0.44/hr
| GPU | GPUs per template | Starting rate |
|---|---|---|
| NVIDIA H200 SXM | 1 · 2 · 4 · 8 | $3.99/hr |
| NVIDIA H100 SXM | 1 · 2 · 4 · 8 | $2.69/hr |
| NVIDIA RTX Pro 6000 | 1 · 2 · 4 · 8 | $1.89/hr |
| NVIDIA A100-80GB | 1 · 2 · 4 | $1.49/hr |
| NVIDIA A100 | 1 · 2 · 4 | $0.89/hr |
| NVIDIA A30 | 1 · 2 · 4 | $0.41/hr |
| NVIDIA L4 | 1 · 2 · 4 · 8 | $0.44/hr |
Storage is billed separately. Confirm the selected region and configuration in the dashboard.
Choose your environment
Templates or GPU VMs?
Choose the level of system control your project needs.
Both options provide GPU compute; the environment and level of control differ.
- Starting point
- Templates
- Configured framework container
- GPU VMs
- Virtual machine with root access
- Development
- Templates
- SSH, notebooks and editor access
- GPU VMs
- Install and manage your own tools
- System control
- Templates
- Packages within the container
- GPU VMs
- OS, kernel and Docker
- Typical fit
- Templates
- Experiments, fine-tuning and visual workflows
- GPU VMs
- Custom system dependencies and infrastructure
| Requirement | Templates | GPU VMs |
|---|---|---|
| Starting point | Configured framework container | Virtual machine with root access |
| Development | SSH, notebooks and editor access | Install and manage your own tools |
| System control | Packages within the container | OS, kernel and Docker |
| Typical fit | Experiments, fine-tuning and visual workflows | Custom system dependencies and infrastructure |
Storage and billing
Keep your work between sessions.
Plan for retained storage as well as the time your GPUs are running.
Running compute
GPU compute is billed by the minute while the instance runs. Pausing compute stops GPU charges.
Retained storage
Storage remains billed while retained, including when compute is paused. The USD storage rate is $0.00014/GB/hour.
Files across instances
Use a network filesystem to reuse data across your own instances in the same region.
Explore Filesystems↗Before you start
Common questions.
Practical details for choosing and using this product.
What is a Template?
A Template is a managed container with a configured framework and GPU software stack. Choose a GPU, launch the environment, and add your code and data.
Can I run Docker or change the kernel?
Use a GPU VM when you need your own Docker daemon, kernel modules or operating-system configuration. Templates provide control within the managed container.
Does a template API scale automatically?
An exposed HTTP service runs on your instance. It does not automatically create workers or scale to zero with requests. Use Serverless for that workflow.
Will pausing delete my files?
Storage is retained across pause and resume and continues to incur charges. Destroying an instance deletes its instance storage; independent network filesystems have their own lifecycle.
Which template should I choose?
Start with PyTorch or TensorFlow for model development, Axolotl for language-model fine-tuning, or an image-workflow template such as ComfyUI. Check the dashboard for the current catalog and launch settings.
Start with a configured environment.
Pick a template and choose the GPU capacity for your next run.