Reference Architecture#

Cosmos Transfer for Synthetic Data Generation#

The synthetic data generation workflow begins with the integration of 3D assets and spatial inputs, which are processed through the Omniverse Kit and CARLA to produce ground-truth videos. These videos, combined with real-world RGB sensor data, are transferred using NVIDIA Cosmos Transfer to create augmented scenes such as changing the weather or road conditions.

These augmented scenes are then curated and used for training AI models and policies. Once trained, the models are deployed across various domains such as autonomous vehicles, robotics, and surveillance systems. The entire pipeline is powered by the NVIDIA RTX PRO Server and Jetson Thor hardware, ensuring high-performance computing for both training and deployment phases.

Synthetic Data Generation Reference Architecture

Bill of Materials#

Best in Class Architecture#

System Specifications for RTX Pro Server

RTX PRO SERVER

8x RTX PRO 6000 Blackwell Server Edition GPUs

800 GB GPU Memory

Virtualization

VMware

Red Hat

Containerization

Kubernetes

Storage

Scalable / Interoperable / Secure

System Specifications for NVIDIA DGX GB300

RTX PRO SERVER

72 NVIDIA Blackwell Ultra GPUs, 36 NVIDIA Grace CPUs

Up to 21 TB GPU Memory

Virtualization

VMware

Red Hat

Containerization

Kubernetes

Storage

Scalable / Interoperable / Secure