NVIDIA Three-Computer Solution

英伟达三台计算机Advanced

NVIDIA's framework splitting robot development into three “computers”: for training, simulation, and deployment.

“Three computers” is the framework NVIDIA uses to promote physical AI, first laid out in an official blog post in October 2024 and updated in August 2025. The first computer is DGX, used to train robot foundation models — for example, post-training GR00T or Cosmos. The second is Omniverse and Cosmos running on RTX PRO servers, paired with Isaac Sim and Isaac Lab, used for simulation, generating synthetic data, and testing policies. The third is a Jetson AGX Thor installed on the robot itself, handling real-time inference and control. The framework strings together “train, simulate, deploy” into one chain, with each link matched to NVIDIA hardware and software; understanding it clarifies NVIDIA's position in embodied AI — it doesn't build complete robots, but supplies the toolchain that everyone else uses.

ExampleA humanoid-robotics team trains a policy on DGX, runs reinforcement learning and testing in Isaac Lab, and then deploys the model to the robot's onboard Jetson Thor — exactly this division of labor.

Also called: Train / Simulate / Deploy Framework
Related: NVIDIA · Physical AI · NVIDIA Jetson Thor · NVIDIA Isaac Sim · NVIDIA Cosmos · Selling Shovels
Sources: Physical AI Accelerated by Three NVIDIA Computers for Robot Training, Simulation and Inference (NVIDIA Blog)
As of: 2025-08