Problem
Currently, devboxes works well for creating isolated development environments, but GPU-based workloads are difficult to use.
This is especially a problem for developers working on AI/ML workloads, such as:
- Local LLM inference
- Model fine-tuning
- CUDA-based development
- GPU accelerated data processing
Today, users need to manually configure GPU passthrough, CUDA runtime, and container settings outside of devboxes, which makes the development environment less reproducible and increases setup complexity.
A native GPU support option would make devboxes much more useful for modern AI development workflows.
Proposed outcome
Add optional GPU support when creating or running a devbox.
Possible usage examples:
devboxes create my-ai-env --gpu
### Alternatives and tradeoffs
_No response_
### Project boundaries
- [x] I considered persistence, security, and Kubernetes portability.
Problem
Currently, devboxes works well for creating isolated development environments, but GPU-based workloads are difficult to use.
This is especially a problem for developers working on AI/ML workloads, such as:
Today, users need to manually configure GPU passthrough, CUDA runtime, and container settings outside of devboxes, which makes the development environment less reproducible and increases setup complexity.
A native GPU support option would make devboxes much more useful for modern AI development workflows.
Proposed outcome
Add optional GPU support when creating or running a devbox.
Possible usage examples: