NVIDIA Open-Sources NVCF: Full GPU Function Platform Now Available
Key Takeaways
- ▸NVCF is now fully open-sourced under Apache 2.0, enabling organizations to run the entire GPU function orchestration platform themselves
- ▸The three-plane architecture (control, invocation, compute) separates concerns and enables independent scaling across the platform
- ▸NATS JetStream serves as the backbone for inter-plane communication, handling request routing, queueing, and multi-cluster coordination
Source:
Summary
NVIDIA has open-sourced the complete NVCF (NVIDIA Cloud Functions) platform under the Apache 2.0 license, making the full infrastructure behind its managed GPU function service publicly available. The release includes the control plane, invocation plane, compute plane, CLIs, Helm charts, and database migrations in a single monorepo at github.com/nvidia/nvcf. The platform uses a three-plane architecture connected through NATS JetStream, enabling independent scaling of control, invocation, and compute layers. Currently, while the source code is fully inspectable under Apache 2.0, the containerized images are distributed via NVIDIA's NGC registry, with ongoing work toward a fully open-source build pipeline.
- The platform powers NVIDIA's managed services including build.nvidia.com and GPU cloud workflows across partners like CoreWeave
- Container images currently require NGC registry access, with the community working toward fully open-source builds
