Getting Started#

Omniverse Streaming on NVCF is a self-hosted Reference Architecture for deploying and scaling streaming applications on NVIDIA Cloud Functions (NVCF). The Reference Architecture provides a defined set of artifacts, versions, and steps that fit together into a validated streaming deployment on AWS or Azure infrastructure. Follow it as documented, or adapt it to your environment.

NVIDIA Cloud Functions (NVCF) is a platform for deploying, managing, and running GPU-accelerated workloads at scale. It routes inference, streaming, and other GPU work to worker clusters, so you can scale demanding workloads with less infrastructure to run yourself. See the NVCF overview for platform details.

Architecture Overview#

Omniverse Streaming on NVCF consists of these key components:

Control Plane

Manages Cloud Function lifecycle, handles invocation requests, and orchestrates workloads across GPU nodes.

GPU Worker Nodes

Execute your containerized Omniverse applications. The NVIDIA Cluster Agent (NVCA) manages workload deployments on these nodes.

Caches

Accelerate function deployments and improve performance for simulation workloads through container, shader, and content caching.

Low Latency Streaming

Enable real-time WebRTC streaming for Omniverse Streaming workloads via reverse proxy VMs.

Kit Development

Build Omniverse Kit applications configured for NVCF streaming and Cloud Function deployment.

Cloud Function Deployment

Containerize your applications, create Cloud Functions, and deploy them to NVCF clusters.

Web Client Development

Create client applications that invoke Cloud Functions and handle streaming connections.

Choose Your Journey#

Select the path that matches your role and objectives:

NVCF Deployment and Operations

For Platform Engineers and SREs

Deploy and operate a self-hosted NVCF stack on Amazon EKS or Azure AKS for running Omniverse workloads.

  • Provision infrastructure and clusters

  • Deploy control plane and GPU workload services

  • Configure caching and streaming components

  • Set up observability and monitoring

  • Validate deployments with performance testing

Self-Hosted Cluster Deployment and Operations
Omniverse Development with NVCF

For Kit and Web Developers

Create, deploy, and integrate NVIDIA Cloud Functions with client applications.

  • Build Kit applications for NVCF

  • Containerize and deploy Cloud Functions

  • Develop web clients with streaming support

  • Integrate with the intermediary proxy

  • Troubleshoot development issues

Omniverse Development for Self-Hosted Deployments

End-to-End Flow#

The following describes how components interact when a user invokes a Cloud Function:

  1. A client application sends a request to the NVCF control plane API

  2. The control plane validates the request and routes it to an available GPU worker

  3. NVCA on the worker node provisions a pod with your containerized Kit application

  4. For streaming workloads, the Low Latency Streaming service establishes a WebRTC connection

  5. The application processes the request and streams results back to the client

  6. Caching layers accelerate subsequent requests by storing containers, shaders, and content

Next Steps#